This week’s lesson was already on its last draft by the time this news was announced, so I feel like I’m extremely aligned with the times. Not to say that all the hype around OpenClaw the past few weeks hasn’t helped my “intuition,” but this just confirms that design talk is now extremely behind.

It’s important to understand the basics of how AI works and the building blocks of LLMs and LLM interactions, but OpenClaw is just the first example of the future of design I’ve been harping on about for the past two years.

The future will be about designing multi-agent experiences.

The question is, are we ready to start designing experiences that might not rely only (or at all) on click/tap interactions? Are the majority of us equipped to work on such experiences? I don’t know.

What I do know is that if we understand the build and technical possibilities of these systems, we will have an easier time innovating and ultimately designing these experiences.

The Anatomy of an AI Agent

IBM Defines an AI Agent as the following:

AI agents are machine learning models that mimic human decision-making to solve problems in real time. In a multi-agent system, each agent performs a specific subtask required to reach the goal and their efforts are coordinated through AI orchestration.

While generative AI models focus on creating content based on learned patterns, agentic AI extends this capability by applying generative outputs toward specific goals. OpenAI’s ChatGPT might produce text, images or code, but an agentic AI system can use that generated content to complete complex tasks autonomously by calling external tools.

Understanding the anatomy of an agent helps design these new experiences.

The Reasoning Engine is the brain of the agent. It takes in context and reasons what to do next, generating a plan. This is where the probabilistic nature we discussed in Week 1 comes in, because the agent’s decision making is uncertain.

The Tools are the agent’s hands. An LLM can only generate text, so to actually do things out there on your device or on the internet, the agent needs access to tools or functions it can call. Searching the web, looking through a database, sending an email, creating a file… these are all possible because it has the ability to access an existing tool that helps it execute these tasks.

Memory is the agent’s… well, memory. Agents need to remember conversations, context, what they’ve done, what they’ve learned, and what the user cares about. There’s short-term memory that focuses on the current task, working memory that focuses on the current state of the plan, and long-term memory that remembers user preferences, conversations, action history, and other accumulated knowledge.

The orchestration loop is the engine that ties it all together. The agent operates in a cycle—first perceiving the current state, then reasoning what to do, taking an action, and finally observing the result, repeating the loop if necessary. This loop runs until the task is complete, the agent gets stuck, or it decides it needs human input.

From a response machine to an orchestration machine

Your current everyday interactions with AI are still very input-output based. You ask ChatGPT or Claude a question and it responds. You probably go back and forth a couple of times before you get to what you need, but the whole time you’re really driving the interaction. Whatever the AI responds with, it’s on you to decide what your next action is.

Example

Let’s say you’re in charge of planning a team offsite somewhere in the Hudson Valley. You can use an AI chat to help you plan by giving it a prompt like “give me accommodation ideas in the Hudson Valley for 12 people.” The chat will give you a list of accommodations which you will then go through to check availability and compare pricing. After doing the research and narrowing down to a few options, you will then coordinate calendars and email/call to make and confirm a reservation.

With a chat, researching options and contextual recommendations have become much easier. However, there’s still a lot of manual work post research that you have to do.

An AI agent works a little differently. A user gives an AI agent a goal and it is on the agent to figure out how to accomplish it. The agent might break the goal into steps, decide which tools to use and install, and after executing on those steps, it evaluates the result and corrects itself if the result is not quite what the user asked for.

Example

Instead of prompting your agent to give you a list of accommodations, you would prompt it to actually plan the team offsite. You might prompt something like “Plan a team offsite in the Hudson Valley for 12 people sometime in March. The budget is $5000. Book the best option.” The agent (or a collection of agents) will then do everything from searching for accommodation and checking availability against your team’s calendars, to drafting a booking request. The agent then comes back with a final recommendation or just books it, depending on its configuration.

Steps and interactions with these types of systems scale back significantly, with the user ultimately going from the driver’s seat to the passenger seat.

How this shifts design priorities

Agentic systems break several core assumptions that traditional UX and even GenAI UX relies on. The future of design will be focused on designing easy delegation, process supervision, error prevention, and execution.

1. You’re no longer designing multi-step flows or a conversation, you’re designing delegation and supervision

In an agentic system, the user gives the agent a goal and the agent takes over from there. As a designer, our thinking now shifts from “What does the user need to be able to do?” to “What does the user need to know?”

We need to understand the agent process and what the user needs to see, know, and allow during that process. We start asking questions like:

• What is the agent allowed to do?

• How much should it do before checking in?

• What happens when it makes a mistake on Step 3?

• How does the user even know what the agent is doing?

Experience design becomes more about what information the user needs to see and when, as well as when the system needs feedback.

2. Control and intervention become the most important interactions

What happens when a user is waiting and watching an agent work, only to realize it’s going in a completely different direction? How do you as a designer decide when the user can or can’t intervene or how much control over the process they have post delegation?

If the agent has already sent an email on your behalf and is now working on step two of booking, you can’t just undo sending an email. This is why a lot of work will go into thinking through different scenarios before feature and flow design, and why we will need to anticipate the worst cases and design for them.

3. Trust needs to be reiterated and built, not assumed

I think the biggest topic in AI is and always will be safety and trust. The doom and gloom predictions always talk about agents going rouge and behind the user’s back. So how do you create experiences that user’s trust?

Distrust of an agent can be managed by helping users develop an accurate mental model of what the agent can and can’t do. New users might need more hand holding through the agent process, seeing more confirmations, more explanations, and more chances to review the agent tasks. Over time, the user can then choose to “trust” the agent more, allowing the interface to progressively reduce confirmations and pauses in the agent process.

4. Designing for error prevention or correction now becomes designing for failure diagnosis

If an agent makes even one wrong assumption during the process it can lead to a domino effect of bad decisions. By the time the user sees the final result, the root cause might be buried several layers deep, and the user is left to clean up the mess IRL.

That’s why designing transparency and diagnostic experiences will be important. Users need to be able to trace back through the agent’s decisions and understand where things went wrong. Typical error messages like “something went wrong” are useless in this context, because the agent doesn’t think anything is wrong. The interface needs to show the chain of reasoning so you can stop it in time or go back and determine where things went wrong.

This Week’s Challenge

Pick a task you do regularly at work or in your personal life. Try thinking of something that takes 30 minutes or more and involves multiple tools or systems.

Map it as an agentic workflow:

1. What is the goal you’d give to an agent?

2. What tools and access would the agent need? (email, calendar, database, files, web search, etc.)

3. Break the process into steps.

   1. Which steps could the agent handle autonomously?

   2. Which ones need your approval?

4. Where are the potential failure points? What happens if the agent gets step 3 or 7 wrong?

Sketch a rough flow and determine where the user sees progress? Where do they and can they intervene? Where do they review results?

This exercise helps bridge understanding agents conceptually and designing for them. Practicing and refining this type of thinking and problem solving separates you as a designer who uses AI from a designer who designs AI systems.

I’m curious to hear about the processes you selected to flow out and reimagine as an agentic experience. Share what you’ve discovered in the comments!

Extra Reading

Additional Resources

Medium — From Products to Systems: The Agentic AI Shift

UX Matters — Designing for Autonomy: UX Principles for Agentic AI

yuj on Medium — Navigating the Agentic Era: Redefining UX for Real-World Impact

Interesting Reads from the Week

Wired — I Loved My OpenClaw AI Agent—Until It Turned on Me

Trending Topics — OpenClaw: Europe Left Peter Steinberger With no Choice but to go to the US

TechCrunch — Have money, will travel: a16z’s hunt for the next European unicorn

For questions, suggestions, collaboration, or consulting projects reach out to Jelena at hello@jelenacolak.com

Going through the basics of AI and machine learning these past few weeks hopefully gave you a better understanding of some of the building blocks of AI. Probabilistic systems, types of machine learning… It’s all pretty standard stuff. Almost all of your daily digital experiences have been made possible by this technology. Anything that recommends, consolidates, and labels things for you is powered by ML.

However, ML has evolved since. The only thing you have to know right now is that there are more advanced subsets of ML that are more sophisticated and can process even more complex patterns than the ones we talked about last week. If you’re curious about it, read more about deep learning to understand what I’m talking about.

Now, what we want to focus on this week is a branch of AI we’ve all become accustomed to using, and that’s Generative AI or GenAI.

As a professional, you will run into ML a lot when working with traditional businesses and institutions. Most of them have just started thinking about how they can use years of collected proprietary data to make their internal (and sometimes external) systems smarter and more useful for people.

The more advanced companies will look for custom AI agents. What that usually means is a custom version of GenAI. The goal is usually to level up processes that already exist in hopes of making them more efficient. A chat that helps sales teams recommend products, a chat that helps determine what audience to target with your next ad… the examples are endless.

In order to make these happen, you have to take an existing LLM and train it with your data. How do we do that?

Well, let’s cover some basic ground first.

What is GenAI and how is it trained?

The name suggests its meaning—Generative AI or GenAI is a broad field of AI that generates new content. It can generate text, images, audio, code… The platforms we collectively refer to as AI platforms or AI Chats, like ChatGPT or Claude, are GenAI platforms.

The foundational engine for many of these platforms is something called a Large Language Model or LLM. It’s a specific type of GenAI that specializes in understanding and generating human-like text and code.

If we learned anything last week it’s that if we want to create a probabilistic system, we need to train it on a certain set of data. Netflix’s recommendation system or Uber’s surge pricing feature are pretty straightforward examples of a simple probabilistic system. So how does one train a model to generate content and responses in plain human langauge?

Stage 1: Pre-Training (Self-Supervised Learning at Massive Scale)

The goal at the beginning is to teach the model to predict the next word. The model is given billions of web pages, books, articles, code repositories, and any other source of text information, and then it learns to predict what word comes next.

Example:

If you show it “The capital of France is ___” it will learn through the billions of text information it has that the next word is most likely “Paris.”

It uses self-supervised learning (a subset of supervised learning) where the system starts labeling the text itself. The next word in every sentence becomes the “correct answer” the model tries to predict.

The result:

At this stage the model becomes incredibly good at pattern recognition. It has looked through so much text it learned everything from grammar and syntax, writing styles and formats, to cultural references and idioms.

However, this just ends up creating another autocorrect, doesn’t it? It might complete questions like “How do I bake a cake?” with “How do I bake a pizza?” instead, because that’s the most common pattern it has seen in all the text it read. That’s not helpful if we’re trying to create an assistant that would need to actually give you information so you can bake a cake or do anything else.

So how do we get it over this hump?

Stage 2: Fine-Tuning (Supervised Learning with Human Demonstrations)

We teach the model what helpful responses look like. Usually, humans write example conversations showing what a good AI assistant would say. This involves thousands of examples and demonstrations to fine tune the model.

Example:

Let’s say the user asks “Explain photosynthesis simply.” A bad response from the system would be something like “Photosynthesis photosynthesis light energy carbon dioxide glucose…” A good response would explain it in everyday language saying something like “Photosynthesis is how plants turn sunlight into food. They use light energy to convert water and carbon dioxide into glucose (sugar) and oxygen...”

To get to that type of response, the model is fine-tuned on these human-written examples.

The result:

At this stage, the model starts to understand it’s having a conversation. It learns how to structure answers, when to ask clarifying questions, appropriate tone for different conversations, etc.

However, we hit a wall at a certain point. You can only write so many example conversations and you can’t predict every possible user input. The model at this stage still makes mistakes and has a tendency to be overconfident while giving unhelpful responses.

Stage 3: RLHF (Reinforcement Learning from Human Feedback)

In this final stage, we teach the model to optimize for what humans actually prefer. Instead of showing the model correct answers, you show it human preferences by:

1. Generating multiple responses to the same prompt

2. Ranking them: “Response A is better than B, B is better than C”

3. Training a “reward model” to predict human preferences

4. Using reinforcement learning to optimize the main model by generating responses that the reward model thinks humans will prefer

Example:

Ever noticed that ChatGPT allows you to thumbs up or thumbs down a response?

That’s a reinforcement learning interaction.

Ever gotten two responses to one question with a prompt to select which answer you like better?

Reinforcement learning.

The feedback you design into your product literally trains the AI. Thumbs up/down, regenerate, edit, and other interactions aren’t just fun UI interactions, they’re the actions keeping the reinforcement learning loop alive.

The result:

The model starts feeling conversational and helpful. At this stage users feel its evolution from a text prediction system to something that feels like an assistant.

How would you train a custom GenAI experience?

You may be thinking “ok great Jelena, but what does that have to do with anything I might do for my clients or for the product I’m working on?”

Well, if you understand how GenAI comes to life, it’s easier for you to understand how to build a custom one. This three stage approach is a helpful approach for when you need to start ideating on what your client wants and needs.

Here’s an example that might put it into perspective.

Case Study: Building a Financial Advisor Assistant

The Business Problem:

A financial services company wants to build an AI assistant that helps financial advisors answer client questions about investment products, retirement planning, and insurance options. The assistant needs to:

• Explain complex financial concepts in simple terms

• Reference specific company products and policies

• Follow strict compliance and regulatory requirements

• Match the company’s communication style and values

• Actually help advisors close more business

Stage 1: Picking an existing model

Most companies use foundation models like Open AI’s GPT or Anthropic’s Claude models as the basis for their GenAI product. In simple terms, think of it as selecting a computer software you will install on your computer (e.g. Windows 98) which comes with certain functionality out-of-the-box. Installing it allows you to use the computer and its out-of-the-box features, but you are also able to customize it by adding your images, downloading files, installing games etc.

After picking out a foundation model your system comes with these “out of the box” features:

• General financial knowledge, like for example knowledge of different retirement savings accounts like a 401(k)

• Ability to explain concepts clearly in natural language

• Understanding of common financial scenarios and questions

• Professional tone and communication skills

Knowing what you have allows you to determine what you do not have but need for your custom product. In this specific scenario, your out-of-the-box model wouldn’t have:

• Knowledge of your company’s specific products

• Your company compliance requirements and regulatory constraints

• Your advisor workflows and internal processes

• Your brand voice and communication guidelines

• Recent market conditions or company-specific data

As a designer, you need to understand first what your custom chat needs to be able to do and what type of questions it should be able to answer. Then, if you understand what the foundation model can already do, you can accurately contribute to scoping what needs custom training. In this example, you don’t need to teach the model what a 401(k) is, but you do need to teach it how the company talks about 401(k)s.

Stage 2: Create the experience

At this stage you would start mapping out what the custom model needs to know. Things like:

• Specific product names and features

• How to position products based on client demographics

• Compliance-approved language

• When to offer the user next steps

• Your company’s tone

You would work on defining what a good conversational experience looks like. I like to utilize service blueprints in this phase, assigning a swimlane to user intent/need, a swimlane for the actual conversational input, and then a swimlane for possible outputs from the chat. While you can’t always design responses word for word, this helps the people testing and fine-tuning the model understand what information absolutely needs to be in each response.

Things you can specify as part of the user experience:

• How detailed should responses be?

• What tone matches our brand?

• When should the assistant ask clarifying questions vs just answer?

• How do we handle topics we can’t advise on?

• Does the chat provide a suggested next action?

• What does “compliance-approved” language actually look like?

Stage 3: Designing the Reinforcement Learning experience

This is where your interaction design chops come in. Remember, we want to make sure the model can learn from reinforcement. This is where you account for specific UI design that consistently helps with reinforcement learning. Adding thumbs down or thumbs up by each answer, a special icon to label compliance violations… There are many ways to incorporate reinforcement feedback seamlessly into an experience.

1. Direct Feedback

You first need to determine what type of feedback the model needs. The interactions can be a general thumbs up or thumbs down, but you can also make it more nuanced:

• Finding ways to label responses as “Too generic” so the model can do better in personalized responses

• Labeling responses as “Too technical” so the model can learn to simplify its responses

• Indicating a product suggested was wrong, so the model can learn to make better suggestions

• Creating ways for users to label something as a compliance issue so the model can learn specific regulatory boundaries

2. Implicit Signals

Indirect feedback can be gathered by various user behaviors. Sometimes they’re the only feedback a system might be receiving, so designing the experience to support these types of interactions is important.

Some frequent implicit feedback includes:

• User copies a response without editing it would indicate a strong positive signal

• User regenerating a response multiple times indicates the model’s current approach isn’t working

• User edits the response before sending it further indicates the response was close to good, but needed refinement

• User immediately closes the tool/chat would indicate that the response was most likely unhelpful and has frustrated the user

3. A/B Testing

As someone that frequently answers UX questions with “it depends,” A/B testing is the best thing since sliced bread. It also allows us as designers and builders to learn what users want instead of assuming things. Some things you can test in these types of experiences:

• Do users prefer bullet points or paragraphs in their responses?

• Do users prefer short responses with follow-up questions or comprehensive answers?

• Do users prefer product centered responses or client centered language?

Why Understanding How GenAI is Trained Matters for Design

Most designers treat generative AI as a black box. Understanding the way AI is trained and what is supported by interactions gives you a strategic advantage in the space.

1. You become aware of what can cause model failure

Understanding pre-training, fine-tuning, and RLHF limitations allows you to help bypass these limitations as much as you can with design.

2. You start to understand what feedback actually does

Understanding what the feedback does allows you to find innovative ways to collect feedback that will help the model learn, but also improve user satisfaction.

3. You understand the extent of ambiguity you can design for

Understanding that generative AI is supervised learning and reinforcement learning helps you become comfortable with designing for variability instead of the consistency we’ve been used to throughout our careers. You can also plan for the evolution and improvement of the product because you understand what that might look like.

UX Homework: Audit the User Experience of GenAI Platforms

Pick any conversational AI product like ChatGPT, Claude, Copilot, customer support bot, etc. and try to identify evidence of reinforcement learning. Look for:

• Design choices that support reinforcement learning

• In conversations with the product, look for clarifying questions from the chat

• Note for any language that admits uncertainty

• Look for instances when the product asks you to rate two different answers

Try to look at these examples and think about how you would change these design decisions, whether they’re conversational or UI.

Extra Reading

Additional Resources

NN/G — The UX of AI: Lessons from Perplexity

IBM Design on Medium — Design Principles for Generative AI Applications

Adobe Design — Designing for Generative AI Experiences

Interesting Reads from the Week

HBR — AI Doesn’t Reduce Work

Matt Shumer on X — Something Big is Happening

For questions, suggestions, collaboration, or consulting projects reach out to Jelena at hello@jelenacolak.com

1. Introduction

2. The Three Machine Learning Styles

   1. Supervised Learning

      1. Definition

      2. How it Works

      3. What it Means for Design

   2. Unsupervised Learning

      1. Definition

      2. How it Works

      3. What it means for design

   3. Semi-supervised learning

      1. Definition

   4. Reinforcement Learning

      1. Definition

      2. How it Works

      3. What it Means for Design

3. Why this Matters to Your Work

4. Homework

5. Extra Resources

   1. Additional reading

   2. Interesting reads from the past week

When ChatGPT first launched I got into a few polarizing discussions around AI with my girlfriends who don’t work in tech. Naturally, a lot of them were skeptical about it—their younger siblings were using it to “cheat” in school and skeptics were warning people of doom and gloom. Most of them were rightfully worried that it would affect our brains and the overall job market. They vowed they would never use it.

One interesting observation I made in all of these conversations was that most of them didn’t realize they’ve already been using AI-powered products for quite some time. Spotify, Netflix, their banking apps, autocorrect on their phones… They’ve been “AI” users for years. When I’d share that with them, they’d be in shock.

AI has a different “ambassador” every decade, each basically being a subset of AI. Right now, the “ambassador” is GenAI, 10 years ago it was topics around machine learning, and in 10 years it will be a different subset. So it’s not at all surprising that it’s not clear to people what AI really is.

So what’s the difference? When we talk about AI today, what are we really talking about?

To simplify it, Artificial Intelligence (AI) is the broad concept of machines mimicking human intelligence to perform tasks. It is a field of computer science dedicated to creating systems capable of performing tasks that typically require human intelligence, such as reasoning, learning from past experiences, problem-solving, and understanding language.

Last week we learned about deterministic and probabilistic systems. We kept it simple, relating deterministic systems to the systems we have been most familiar with the past 20-30 years, like your desktop computer or a static website. We related AI to probabilistic systems, systems that can personalize experiences and provide contextual answers to questions, because the AI most of us use on a daily basis is inherently probabilistic.

Some AI, like expert systems in the medical field, rely on determinism more than others. However, probabilistic AI uses statistical models to analyze data, identify patterns, and make predictions, instead of relying on fixed rules. This type of AI is able to “learn” from data and improve its performance over time.

These types of changes in the backend are directly influencing our jobs as designers. Now more than ever we have to understand the relationship between frontend and backend in order to create effective experiences. So how do machines learn?

The Three Machine Learning Styles

1. Supervised Learning: Learning by Example

Definition

Supervised learning is a machine learning technique that uses labeled data sets to train AI models to identify the underlying patterns and relationships. The goal of the learning process is to create a model that can predict correct outputs on new real-world data.1

In order for the model to correctly execute a certain action, you first show it thousands of examples that it can learn from and then make “educated guesses” in the future.

How it Works

Let’s say you’re training your system to automatically filter email spam. You will give it tons of information on emails that have been marked “spam” or “not spam.” The machine will “study” these labeled examples and learn to make predictions on new, unlabeled data.

This is how most of them learn to make those educated guesses. They study labeled examples until they can predict labels for new data.

Remember the probabilistic examples from last week? Some of them are prime examples of systems using supervised learning:

• Uber’s surge pricing is most likely trained on historical ride data, labeled with demand levels and corresponding optimal prices. It learns from past surge events such as concerts, conferences, weather patterns, etc. helping it determine future pricing in instances where demand for rides spikes.

• LinkedIn connection suggestions gets trained on millions of examples where it knows “these people connected” vs “these people didn’t connect.

• Autocorrect is trained on datasets where it knows what people meant to type (the label) vs. what they actually typed. Still weird people are meant to type duck…

• Package delivery estimates are trained on historical delivery data labeled with actual delivery times. It learns patterns from weather, distance, time of year, route complexity, etc. to estimate delivery windows.

What it Means for Design

Knowing that your product and experience is powered by supervised learning helps you make design decisions that allow the system to support an experience with the things it learned.

You have two jobs as a designer:

1. Understand what experience the system’s knowledge is powering.

2. Understand what data the system is collecting for future learnings.

The design needs to support the experience, like in our email example where a spam folder needs to exist so the emails the system labeled as spam are still accessible to the user for viewing but organized in a separate folder. However, if your system is also gathering data for future supervised learning, it should also support that. So, giving users the ability to label their own spam through the UI allows the system to collect new data that’s labeled as “spam.”

2. Unsupervised Learning: The Pattern Hunter

Definition

Unsupervised learning uses machine learning algorithms to analyze and cluster unlabeled data sets. These algorithms discover hidden patterns or data groupings without the need for human intervention.2

You give the system unlabeled data and tell it to find patterns. It clusters similar things together, spots outliers, and discovers hidden structures and connections. Unlike supervised learning, there’s no “correct answer list” for it to learn from. The system just looks for structure in whatever data they have access to, making guesses about what patterns might be meaningful.

How it Works

Let’s say you have a restaurant app and want to create a curated list of restaurants with niche names like “bougie girls brunch” or “NFL brewskis for the boys.” You would have your model learn through unsupervised learning. The system would look at user patterns of restaurant selections, times and days of booking, types of bookings, etc. and cluster selections together. Later, when another user selects a certain restaurant, they will get suggestions of related restaurants from that cluster, even if the restaurants might not serve the same cuisine as your original selection.

Examples from last week that use unsupervised learning:

• Spotify Discover Weekly uses collaborative filtering and clustering, by finding patterns and clusters in listening behavior without being told “these songs go together.” It discovers taste neighborhoods from unlabeled data.

• YouTube doesn’t manually categorize every video into topics like “Lo-fi Hip Hop” or “Mechanical Keyboard ASMR.” It clusters videos based on who watches them together, creating hyper-specific categories that emerge from viewing patterns.

• Your bank groups transactions into categories like “Dining,” “Entertainment,” “Travel” by finding patterns in merchant names and transaction types, not because someone labeled every Starbucks purchase as “dining.”

• Amazon’s “Frequently Bought Together” discovers purchasing clusters by analyzing millions of shopping carts and finding items that appear together more often than random chance would predict.

What it Means for Design

Unsupervised learning surfaces insights humans might miss, but those insights aren’t always meaningful. The system will find patterns, but it is your job to translate algorithmic patterns into human-relevant categories and make sure the platform UI can handle changes without the UX changing.

This is where a good understanding of effective information architecture comes in. You need to test whether discovered patterns actually help users accomplish their goal and then design interfaces that let users easily access and trust these groupings.

3. Semi-Supervised Learning: The Hybrid Approach

The name says it all but here’s the definition

Semi-supervised learning is a branch of machine learning that combines supervised and unsupervised learning by using both labeled and unlabeled data to train AI models for classification and regression tasks.3

Many platforms and systems combine the two and update as they grow. In design terms, understanding where your product lies and what the product needs vs. what the user needs is essential.

4. Reinforcement Learning: The Trial-and-Error Student

Definition

Reinforcement learning is a type of machine learning process in which autonomous agents learn to make decisions by interacting with their environment.4

The system tries different strategies, sees what happens, and gradually shifts its behavior toward whatever gets rewarded.

How it Works

Netflix is a great example of a system that utilizes reinforcement learning. Let’s say you’re on your Netflix homepage looking for what to watch. The system recommends titles based on what you’ve watched and rated in the past (supervised learning), but then it tests which images and arrangements get you to click certain titles, learning over time what works and what doesn’t. That’s reinforcement learning.

Another example of reinforcement learning is dynamic ad bidding. Advertising platforms learn how much to bid on ad placements by trying different bid amounts and getting rewarded when ads lead to conversions. The system explores what happens if it bids higher and analyzes results. It optimizes spending over time through trial and error.

What it Means for Design:

Reinforcement learning means the system is goal-oriented and it optimizes for whatever reward signal you define. This can be a powerful experience enhancer but can also very easily make things go to shit.

If you reward a customer service chatbot for closing conversations quickly, it’ll learn to give short, unhelpful answers. If you reward it for customer satisfaction scores, it might learn to be so agreeable it never enforces policies. The reward function IS the product behavior.

Your job is to define success metrics that actually align with user needs. You need to design feedback mechanisms that give the system good reward signals and also build safeguards so the system can’t optimize itself into problematic behavior. This is where error prevention is crucial. Being very clear on what the product is optimizing and how that could backfire will help build and design a successful reinforcement learning product.

Why This Matters for Your Work

More and more experiences are about personalization, which increases the need to create dynamic experiences, instead of static. When we design AI experiences we have to become very familiar with the relationship between frontend and backend and be clear on what interactions the system requires.

Design discovery questions should include:

• Which ML approach are we using?

• What training data exists or needs to be created?

• What is the long term goal for the system?

• How will the system improve over time?

• What user feedback is needed to do that?

• What could go wrong if the system optimizes incorrectly?

Understanding what you’re working with helps you set requirements for the experience you’re designing. If you know the product needs a dashboard that will have different personalized categories of recommendations, what type of design will handle that effectively? What character limits will you set for the category names? How will the categories be represented visually? UX and UI requirements are shifting from linear journeys towards accommodating dynamic, contextual journeys.

UX Homework: Identifying machine learning types

The exercise this week is something you can do on your commute to work or during your breaks.

Open your phone and take a look at the apps you use daily. For each one that feels “intelligent” try to determine if it utilizes:

1. Supervised learning - making predictions from labeled examples

2. Unsupervised learning - finding patterns in your data

3. Reinforcement learning - optimizing results or the whole system through trial and error

Once you identify the ML type, ask: How does this influence the user experience? Where are the feedback loops? What’s being optimized? How could it fail?

The goal is not to become an ML engineer, but to develop pattern recognition for how systems learn so you can design better interactions around their learning processes and the experiences they can power.

Extra Reading

Additional Resources

IBM — The 2026 Guide to Machine Learning

UX Magazine — The Future of UX Design: How AI and Machine Learning Are Changing the Way We Design

Medium — What is Machine Learning + UX

Interesting Reads from the Week

Forbes — Sam Altman Reveals Why OpenAI Is Poised To Make The Biggest Business Bets Ever

The Verge

— Humans are infiltrating the social network for AI Bots

— Firefox is adding a switch to turn AI features off

Next week we will dive deeper into the evolution of AI and what makes each generation different.

For questions, suggestions, collaboration, or consulting projects reach out to Jelena at hello@jelenacolak.com

1. Introduction

2. Why designers must understand system behavior before interfaces

3. Deterministic Systems

   1. Definition

   2. How deterministic systems are built

   3. Examples of deterministic systems

4. Probabilistic Systems

   1. Definition

   2. How probabilistic systems are built

   3. Examples of probabilistic systems

5. Why this distinction breaks UX patterns

6. Homework

7. Extra Resources

   1. Additional reading

   2. Interesting reads from the past week

One night a few weeks ago, sometime around dinner time, I turned my once-every-few-days episode of Stranger Things (I’ve been rewatching the series since the finale aired). The episode that day happened to be the one where Joyce, Hopper, Mike, and Bob are trying to get themselves and Will out of a demodog flooded Hawkins Lab. In a heroic effort, Bob Newby SUPERHERO runs down to the basement to turn the power back on and reset the building security system by using BASIC.

If you’re not familiar, BASIC is a family of general-purpose, high-level programming languages designed for ease of use. “Ease of use” meaning ease of use for the times — to use some of the first desktop computers (microcomputers) you had to type in code, and basic was “easy.”

This reminder sent me down a rabbit hole of research. We all know the stories of Steve Jobs and his early UI innovations. The Macintosh was the first commercially successful desktop computer with a mouse and a GUI. Yes, it was the first instance of the modern GUI, but the words commercially successful matter, because the first GUI was actually introduced 10 years earlier.

In 1973, Xerox PARC released the Xerox Alto. It came with a steep price - $32,000 (equivalent to $140,000 in 2024) so it didn’t find much commercial success, as you can imagine. However, the Alto was the first computer system to introduce graphical elements that we still use to this day - a bit-mapped display, windows, menus, radio buttons, check boxes, and even a WYSIWYG text editor.

What’s wild to think about is that the people who came up with these elements literally had to come up with them out of thin air. Sure, the elements are simple and relate back to the real world which makes them easily understandable. But the first UX designers weren’t graphic designers, they were engineers. They innovated by understanding the system, not following a list of “best design practices.” Those didn’t exist.

We’ve been designing for the UI logic set in the 70s and 80s ever since. The Xerox team conceptualized it, the Apple teams set the standard. But today, for the first time since, we have an opportunity to innovate again.

Why Designers Must Understand System Behavior Before Interfaces

For decades, designers have been trained to work with deterministic systems.

UI and UX elements like buttons, forms, menus, error states, confirmations, and feedback loops were all invented for software that behaves in predictable, rule-based ways. You perform an action and the system responds in a straightforward and consistent way. E.g., if you click an X button you know that a window or tab will close. Every time you click it, again and again it will close. It follows the interaction logic from the BASIC times. Input, output, input, output. Over time, users learn this behavior with all elements and develop trust.

Artificial intelligence systems do not function this way. They violate this assumption at a foundational level because AI systems are probabilistic, not deterministic.

AI systems do not execute predefined rules, rather, their output is based on patterns learned from data. This completely reshapes what good user experience means and is the reason a lot of “bad UX” is prevalent in current AI products. We’re using interaction paradigms that promise deterministic behavior but deliver probabilistic instead.

That’s why in order to design good experiences for AI interactions, designers should first understand what kind of system they are designing for.

Deterministic Systems: Software as Explicit Instruction

Definition

A deterministic system is a system in which every output is explicitly determined by a set of predefined rules. It’s defined by its ability to produce the exact same output from the same input, every time it is executed, regardless of environmental factors.

Think of it this way:

Deterministic systems are like following a recipe. If you use the same ingredients in the same amounts and follow the same steps, you’ll get the same result every single time. There’s no randomness involved.

How Deterministic Systems Are Built

Deterministic software is constructed through explicit logic written by engineers. These systems rely on conditional statements, procedural flows, and clearly defined state transitions.

At a technical level, deterministic systems are composed of:

• Conditional logic (if/then/else)

  • IF order total is over $50, THEN free shipping, ELSE charge $5.99.

• Boolean operations (true/false)

  • Password matches database?
    TRUE = access granted,
    FALSE = “incorrect password” message

• Fixed algorithms

  • Google Maps route calculation uses the same algorithm to always find the shortest path between two points given current traffic data.

• Explicit error handling

  • If you try to access a page that doesn’t exist you will get a specific “Page Not Found” page instead of the site crashing.

• Predefined constraints

  • If you try to write more than the 280 characters allowed on Twitter, the system won’t let you post.

The system does not generalize or interpret intent. It does not function beyond what has been coded and it has been “told” to do. If an unfamiliar situation arises for the system, it either:

1. Fails

2. Produces an error

3. Returns no result

This is by design.

Examples of Deterministic Systems

A light switch - Flip it up and the light will turn on. Flip it down, the light will turn off.

Submitting a form - If you fill out all the required fields and click the submit button, the form will submit. Every time you stumble upon a form, you’ll expect the same outcome for the correct input.

A combination lock - Let’s say you set the combination to 25-17-39. Every time you enter those numbers again, the lock will open. If you enter the wrong combination, it will stay locked.

Authentication systems -In order to log into your email account you need to enter your username and password. Assuming you entered the correct information, you will be logged into your email.

Probabilistic Systems: Software as Inference

Definition

A probabilistic system is a system that produces outputs based on statistical reasoning rather than predefined rules. Instead of executing explicit instructions, the system estimates the most likely outcome based on the input it received and its learned patterns.

Think of it this way:

Probabilistic systems are like rolling dice. Even though you’re doing the same action (rolling), you can’t predict the exact outcome. You know the possible outcomes (1 through 6) and in theory their likelihood, but each roll could be different.

How Probabilistic Systems Are Built

Probabilistic systems are created through training, not traditional programming.

Rather than writing rules like “If the user types X, return Y,” engineers need to provide the system with large data sets, model architectures, and optimization objectives. The system learns relationships between inputs and outputs by identifying patterns in the data it was provided with.

At no point does the model “understand” meaning of this data in the way humans would find “meaning,” it just learns statistical correlations.

At a technical level, probabilistic systems are composed of:

• Statistical models

  • Spotify’s Discover Weekly analyzes patterns across millions of users and uses statistical clustering to predict what you’ll enjoy.

• Probability distributions

  • Uber’s surge pricing system calculates distribution of likely demand at a certain time (e.g. after a concert) and adjusts prices based on probability curves, not exact rider counts.

• Weighted variables

  • LinkedIn considers a combination of multiple factors like mutual connections, same company, similar job titles, location and weighs them to suggest users you should connect with.

• Training data & pattern recognition

  • Autocorrect on your phone is trained on massive text datasets. It sees you misspell “teh” and recognizes the pattern matches “the” 99.9% of the time based on historical corrections. Who keeps writing duck…

• Confidence intervals & uncertainty margins

  • Package delivery estimates are always given in windows of time based on historical delivery time variance and current variables because the system can’t guarantee an exact day.

This means:

1. There is no single correct output

2. Outputs exist on a probability distribution

3. Multiple responses may be equally valid

4. The system’s confidence is implicit, not guaranteed

Examples of Probabilistic Systems

A slot machine - if you pull the lever with the same force, same time of day, you will get different results each time.

Netflix recommendations - log into your account today and tomorrow at the same time and your “Recommended for You” might show the same titles with different cover pictures, show them in a different order, or even show you completely different titles.

Shuffling a deck of cards - shuffle the same deck of cards twice and you’ll get two completely different arrangements every time.

Social media feed ordering - every time you refresh Instagram, your feed order changes. Every time you do this, the algorithm is deciding what to show you based on your probability of engagement.

Getting a green light at an intersection - if you arrive at the same intersection at the exact same time of day on different days, the light might be red, yellow, or green depending on traffic patterns.

Autocorrect - If you type something like “I’m going to the…,” the next word suggestion will never be exactly the same. It changes every time based on your history, context, and probability models.

Why This Distinction Breaks Traditional UX Patterns

From the user’s point of view, deterministic systems create a stable mental model. Users learn that on any new system, actions are learnable and repetition leads to mastery because outcomes can be anticipated. And if you make a mistake, you can easily correct it by clicking an “Undo” button.

On the other hand, probabilistic systems feel flexible, adaptable, and currently mostly conversational. But this masks a lot of uncertainty, and a system may sound confident even when it is wrong. Unlike deterministic software, probabilistic systems do not naturally expose failure states. Those errors are one of the central design challenges of AI.

Traditional UX patterns work for predictability. These rules came out of observation of predictable user behavior towards the predictable system behavior. Making buttons “easy to find,” designing intentional error messages that appear only when a certain system rule is violated, these are all design rules that have been created through years of system consistency.

The key design assumption in deterministic software is that if a user performs the same action twice, they should get the same result twice. While most UX patterns we’re familiar with rely on guarantees, probabilistic systems don’t offer that.

This Week’s Homework: Redesigning for Probability

Now that you understand the difference between deterministic and probabilistic systems, it’s time to apply it.

Choose ONE (or all, if you’re feeling ambitious) of the three scenarios below. Each represents a real AI design challenge where traditional UX assumptions start to break. Create a design solution that acknowledges the probabilistic reality instead of fighting it.

Scenario 1: The Consistency Assumption

The Problem: You’re designing an AI writing assistant for a legal tech company that lawyers use to draft contract clauses. They’ve reported that “the AI keeps changing how it formats indemnification clauses.” Sometimes it’s a single paragraph, sometimes it’s broken into sub-clauses, sometimes it includes examples.

A traditional UX expectation is that consistency builds trust. But, in AI systems, consistency is not guaranteed and it’s not always wanted. Instead, coherence replaces consistency. The content is coherent (legally sound indemnification language), just not identical.

Your Challenge: Design an interface solution that reframes the expectation from:

“This will generate the exact same clause every time”

to:

“This will generate legally appropriate indemnification language that achieves your intent”

What UI elements, language, or interaction patterns help users trust coherence over consistency?

Scenario 2: The Binary State Assumption

The Problem: You’re designing a customer support AI chatbot for an e-commerce platform. Currently, the system shows only two states:

• Green checkmark = “Issue resolved”

• Red X = “Error — contact human support”

Users are reporting frustration because the AI often:

• Answers part of their question correctly

• Provides a helpful workaround, but not the solution they wanted

• Gives an answer that’s plausible but needs verification

• Understands their question but lacks authority to act on it

None of these fit under either “resolved” or “error.” Deterministic systems have clear success and failure states, while probabilistic systems operate in gradients.

Your Challenge: Design a status/feedback system that expresses the reality of AI responses. How do you show:

• Partial success?

• “Correct under one interpretation”?

• “Helpful but incomplete”?

• Confidence levels?

What can replace the binary success/error pattern?

Scenario 3: The Boundary Assumption

The Problem: You’re designing an AI-powered medical symptom checker. Unlike traditional form-based symptom checkers (which only accept predefined inputs like dropdowns), your AI accepts natural language.

This means users can type literally anything:

• “my left knee hurts when i run”

• “is this mole cancer”

• “my dog ate chocolate what do i do”

• “cure for immortality”

The AI attempts to respond to all of these, even when it shouldn’t. Deterministic systems enforce boundaries by rejecting invalid input. Probabilistic systems attempt to respond to nearly anything.

Your Challenge: Design guardrails that compensate for the AI’s lack of self-regulation. How do you:

• Prevent the AI from answering questions it shouldn’t?

• Redirect users when they’re outside the system’s scope?

• Signal when an answer might be a confident hallucination?

• Acknowledge the user’s input without providing dangerous advice?

What UI patterns create boundaries the AI itself cannot enforce?

Deliverable Format

Sketch it out, create a wireframe, or just write out your thoughts. If you’re up for a discussion, send your solution to me or drop it in the comments and let’s talk! I would love to see what you come up with.

Extra Reading

Additional Resources

NVIDIA — What Is Artificial Intelligence?

Towards Data Science — Deterministic vs. Probabilistic Deep Learning

The Art of Unix Usability — Chapter 2: A Brief History of User Interfaces

Medium — AI in Product Design Concepts: A Primer on Deterministic vs Probabilistic Systems

Interesting Reads from the Week

AI News

— Enterprise AI Adoption Shits to Agentic Systems

— Anthropic Selected to Build Government AI Assistant Pilot

WIRED — Why AI Breaks Bad

Tech Crunch — OpenAI launches Prism, a new AI workspace for scientists

Yahoo!Finance — Phoebe Gates and Sophia Kianni’s Phia raises $35M to ‘make shopping fun again’

Next week we will dive deeper into machine learning - understanding types of learning, when each is used, and how that affects user experience.

For questions, suggestions, collaboration, or consulting projects reach out to Jelena at hello@jelenacolak.com

Not proud because, well, catching yourself scrolling on TikTok for hours on end is not a thing to be proud of. On the other hand proud, because of said scrolling I am often mistaken to be part of the Gen-Z population. In translation, I’m in tune with what’s happening in culture and understand the language of a completely different demographic, which is extremely important in marketing and business.

I see how different the reality of some of my 30+ friends who refuse to download TikTok is different than mine. Sometimes it feels like we live in two completely different countries, have access to completely different musical top charts, and are completely different generations. That’s a topic to unpack for another time, but you get the gist.

Seems like those anti-TikTok friends of mine have always been onto something, though. While we were worrying about how much the app knows us and our preferences and whether that’s good or bad, there have been undetected tech thieves lurking on something else - our likeness.

In the past few weeks, creators have been sharing experiences of their face and voice being stolen. Questionable companies selling questionable products are manipulating creator videos to make it seem like those creators are selling their products. These videos are actually pretty convincing, so much that even the most trained eyes would not think twice about it.

A New Era of IP Infringement

The first video I stumbled upon was by creator and podcaster Arielle Lorre. She shared how beauty company Skaind leveraged existing content from her podcast and digitally manipulated it to make it seem like she was talking about Skaind products.

@ariellelorreTikTok attorneys and legal experts - please weigh in!

Tiktok failed to load.

Enable 3rd party cookies or use another browser

In efforts to have the ad removed, Arielle first messaged the company asking them to take the video down.

Skaind responded, first by apologizing but then stating:

“Our marketing team accessed this content through an artificial intelligence platform without being aware it was a recognized person or person with image rights.”

There are a couple red flags here.

First, the company clearly knew what they were doing because in the video, besides having the AI podcast version of Arielle, they are reusing her skincare videos and other lifestyle content from her TikTok account. Claiming they weren’t aware “it was a recognized person” is clearly a lie. Not to mention that the “interviewer” you can see in the deepfake video is also a popular creator and podcaster Rich Roll, which - you guessed it - also had his original videos tinkered for this ad.

Second red flag is the fact that after this interaction, Arielle of course sent a seize and desist letter to Skaind (and they promptly blocked her, of course) and reported the ad to Meta. However, Instagram reviewed the report and responded back with “this post does not go against our Community Standards.”

Ummmmmm…. what do you mean does not go against our Community Standards?

This irked me, so I went digging.

First statement I found around this topic in Instagram’s Help Center was:

“Under Instagram’s Terms of Use and Meta's Community Standards you can only post content to Instagram or Threads that doesn’t violate someone else's intellectual property rights.”

Interpreted very literally and broadly, that would mean you can’t just use someone else’s videos, tinker with them, and then post as your own.

I went to read further.

“It's possible to infringe someone else's copyright when you post their content on Instagram or Threads, or facilitate copyright infringement, even if you:

• Bought or downloaded the content (example: a song from iTunes)

• Recorded the content onto your own recording device (examples: a song playing in the background during a party, concert, sporting event, wedding, etc.)

• Gave credit to the copyright owner

• Included a disclaimer that you didn’t intend to infringe copyright

• Didn’t intend to profit from it

• Modified the work or added your own original material to it

• Found the content available on the internet

• Saw that others posted the same content as well

• Think that the use is a fair use

• Are using an unauthorized streaming device or service (examples: a “jailbroken” or “loaded” app or service)”

I’m not a lawyer, so if you are, correct me if I’m wrong. I would say this means that even if you somehow got legal permission to use their content, checked all the boxes to make sure it’s not “stealing,” it doesn’t mean you’re not violating someone else’ IP rights.

I continued to read a little further and think hard about this, but suddenly I paused. Huh… well, aren’t we then technically infringing on other’s IP rights almost on a daily basis?

Reposting that Pinterest inspo quote you liked so much, sharing a video on your stories from the latest BRAT tour, tweaking a picture to create a meme… that can all be considered a violation of intellectual property rights.

The internet is basically a black hole of stolen and tinkered IP. Since the start of the social media era, we have all essentially infringed on someone’s IP thousands of times. Most of the time, however, it’s nothing of note to warrant lawsuits or reporting on these platforms. That’s the issue, though. Where does the thin line start and where does it end?

Are people outraged about Arielle’s situation because we’re just not used to this type of IP infringement yet? Will the new era of memes be AI deepfakes? Will our new quote sharing actually be a clip we cut from Mel Robbins’ podcast and posted on our feeds aesthetically, instead of the current Pinterest/Tumbler-like image reshare?

Obviously I’m playing a little bit of devil’s advocate here. It’s my job to question things, remember? I’m a UX Designer and a brat at my core after all. It’s an interesting topic of discussion though… the future of content. A frequent one people, creators, the media, and various experts are having.

“Good artists copy; great artists steal” is a famous saying you’ll overhear the creative director types quoting to their buddy next to you while you’re in line at Cafe Lyria waiting to order your coffee. However, in 2025 it’s true more than ever. Everyone seems to be stealing and we’re all used to it.

Followers maketh man

This brings me to my third red flag. Going back to Skaind’s response to Arielle, the phrases “recognized person” and “person with image rights” kind of give me the heebie jeebies. So now, the right to use my likeness depends on what? The amount of followers I have or whether or not I’m a public figure? If I have image rights to the video, you can’t use it. Buuut, if I don’t, you can do whatever you want with it, including using my face to sell and endorse products?

Small creators and non-creators have been the one to suffer the most when it comes to businesses and other influential people stealing their ideas and art for decades, if not centuries. While it’s nothing new, it’s the sad reality that if you have even a tiny amount of “influence” vs none, you’re more likely to be able to win your case against misuse of your IP or your image.

I’m worried that in this new era of AI Deepfakes and IP lawsuits, the gap will just continue to grow larger and the consequences unbearable. Having your face photoshopped on an Ad is one thing, CREATING A WHOLE VIDEO OF YOU TALKING is a whole other can of worms… which needs to be opened and discarded ASAP.

A Marketplace of Faces

Final red flag, which when I was first watching Arielle’s video went over my head, but in the process of writing this letter burned my eyes, is Skaind’s mention of an “artificial intelligence platform.” Allegedly, their team found Arielle’s videos on, I’m assuming, some type of content marketplace or deepfake platform and used it thinking it was labeled as “free use.” If this is true, and not another lie to cover their asses, then that’s even more concerning.

Why are these videos readily available on these platforms if Arielle didn’t grant them those rights? Has she granted the rights without realizing? I doubt it, her lawyers would have caught it by now and she probably would have included that information in her original video.

So this begs the question, how do we solve this? How do we make sure our videos aren’t out there in the ether being sold for use in not just Ads but questionable, reputation ruining content? This was “just an ad” but what if it was… a revenge porn video? Someone posing as her on dating apps? This can get so ugly, so quickly, and we’re barely doing anything about it.

What happened to Arielle and countless of other creators and non-creators is not okay and should be regulated.

How do we regulate it?

Now that’s a discussion I’d like to have not just once, but frequently.

Are these discussions even happening?

Barely, or at least I haven’t found much about it on the internet. Everything is still very surface level and I have an inkling nothing will be done about it until there’s a situation that warrants someone testifying in front of Congress.

1Every year towards the end of January, a flood of “January feels like a century” memes grace our small screens. “Happy January 75th” or “January was a tough year but we made it” and similar, are relatable, funny, and expected. What I didn’t expect was that in the AI world, the first month of 2025 was literally going to have a year’s worth of drama.

The most notable drama was the introduction of DeepSeek’s new R1 model. The launch shook Silicon Valley and had the stock market shivering, with Nvidia stocks experiencing a 17% drop. However, the unraveling of this whole situation is giving Cady Heron arrives at North Shore High School.

Not long after DeepSeek’s downloads skyrocketed, big tech CEOs started commenting on the “newcomer,” from OpenAI’s Sam Altman describing it as “clearly a great model” to Apple’s Tim Cook sharing a very on brand response: “innovation that drives efficiency is a good thing.” Definitely approved by Apple’s Legal and PR departments.

Have the The Plastics (AI industry equivalent name suggestions welcome) invited DeepSeek to have lunch with them for the rest of the week? What days do they wear pink?

Jokes aside, it was definitely a wakeup call for Silicon Valley, which expedited some launches and deals, like OpenAI’s introduction of Deep Research2 and announcing their deal with South Korean Fintech giant Kakao. Who are actually Cady, Janis,3 and Damian in this situation, and who are The Plastics? Time will tell.

From a consumer perspective, whether you are a super knowledgable AI/ML expert or a newbie just starting to dabble in GenAI, it’s hard to keep track. ChatGPT, Claude, Gemini, Llama, DeepSeek’s R1… the list of platforms, agents, apps, you name it, just keeps getting longer and longer. It’s hard to discern what’s great for what, who owns what, who should be using what… So, in the spirit of this very girl world-like chain of events - where CEOs are giving each other compliments yet scheming behind each other’s backs, and keeping their friends close but enemies closer - I’ve decided to share the ultimate who’s who in school guide to all the big GenAI tools. Do with this information what you will. It’s up to you to decide who you’ll be sitting with.

The Ultimate Tech Brat Guide to AI Tools

In no particular order.

Levels

Freshman - Complete beginner users
Sophomore - Experienced beginner users
Junior - Intermediate users
Senior - Advanced users

ChatGPT

Access Link

The lowdown:

ChatGPT is an AI assistant offered in chat form with various interaction capabilities, including text, images, code, and voice. One of the original leaders in the GenAI and AI revolution.

Quick facts: Her dad is Sam Altman, the CEO of OpenAI. He’s like everywhere. She’s friendly to Freshmen, though.

Sit with her if:

• You need assistance with everyday school and work tasks like writing4, coding, research, etc.

• You need help with image analysis, document parsing, and similar vision-based tasks.

• Your company has Microsoft Copilot integration and promotes using it internally.

• You have moxie and fancy yourself a custom AI you can develop around a certain task, topic, and similar.

Why she’s cool:

I like to use ChatGPT for planning and brainstorming strategies around work, my career, or my personal projects. I also like to use it as a research assistant, particularly when I need to compile a list of sources or informational items. Since ChatGPT is connected to the internet it can search through latest information quickly. For example, for this list I used it to compile a list of ChatGPT competitors just in case I forgot one or wasn’t aware something existed.

What’s making her not cool:

As a UX designer, when I compare it to some other competitors, they lack some support language and labels. You’ll see what I mean when I talk about Claude. Its responses also sometimes feel lazy.

Survival Tip:

The more you use it, the more it will learn about you and can reference what you “talked” about in previous chats. This is helpful when you’re working on something for a long time and keep coming back to the platform to do some research or solve problems.

Gemini

Access Link to Chat, also integrated in your Workspace apps

The lowdown:

Gemini is an AI assistant integrated into Google Search, Workspace, and Android, but is also offered in chat form.

Quick facts: She’s part of the Google family, which is basically like royalty. She’s also Freshman friendly though, but you’ll get along better if you’re a Sophomore or Junior.

Sit with her if:

• Your work platform is Google workspace and you spend a lot of time using Google Docs, Gmail, Sheets and similar.

• You want to make your workflows more efficient and are open to changing your habits (in Google workspace apps).

• You want to use it just like ChatGPT but have access to Google’s search engine and not Bing (which is what ChatGPT is integrated with)

Why she’s cool:

My favorite Gemini feature is actually the AI overview feature that’s part of Google search results now. I like that it summarizes some key information that it is pulling from various sources on the internet AND it also links to the sources so you can search on your own. In terms of using Gemini in my workflows, I’ve been playing around with it, but I don’t use Workspace apps enough to give a thorough review yet. Will definitely report back.

What’s making her not cool:

You know I talk a lot about how AI implementation in existing platforms is still lagging in terms of experience innovation. There’s opportunity for Google to rely less on prompts and to use some of their innovative genius (if they still possess any) and reimagine their experience which will be powered by Gemini. Why do I have to consciously decide to use Gemini? Just make my life easier since you already have the tech to do so. I’ve also noticed it has a tendency to hallucinate5, which is problematic if your goal as a company is to have people rely on it in their workflows.

Claude

Access Link

The lowdown:

Claude is an AI assistant that’s also offered in Chat form, but focuses a lot on information privacy and safety. This means that Claude is also not connected to the internet, so their model’s “knowledge” is capped to the date it was released/updated. They have a strong Enterprise focus and are favored by legal, HR, healthcare, and similar teams.

Quick facts: Her parents are Anthropic. They’re like super strict about security, but that’s made her so nice to Freshmen.

Sit with her if:

• You’re a complete beginner and couldn’t figure out ChatGPT.

• You need thorough problem solving and assistance in writing, coding, and other projects.

• You have long form content you need help summarizing. For example, a large research paper.

• You’ve gotten used to GenAI in general and now have specific types of projects you would like to work on, but you don’t need to rely on information from the internet.

• You work on projects with a lot of IP or sensitive information.

Why she’s cool:

Claude is my favorite AI assistant. I’ve been using it more than ChatGPT for the past few months and the only instance when I don’t is when I need to pull information from the internet. First thing I loved about Claude was how they really thought through the details that make a big impact in the overall user experience. I really geek out about that kind of stuff. For example, the first thing you notice when you start using Claude is the two options right in the prompt box:

When you click on the first dropdown, the system will show you different Claude model options you can choose from.

Choosing between models is nothing revolutionary because other AI assistants offer the same, but the difference is that Claude gives you a little description of each model and the recommended use case. This is clearly an effort to welcome any level of user to Claude.

The second dropdown saves you time in specifying what type of response you need.

You can even create and edit styles, which comes in handy if you regularly use the platform and want it to learn different styles of writing responses.

Maybe you work in marketing and the brand you write copy for has a specific brand voice. Claude can help you with that.

Don’t get me wrong, other AI assistants can help you with that as well, but Claude’s experience makes it easier for the everyday user and does what I keep on preaching - uses AI to enhance the experience and not just rely on prompts.

Other than geeking out about their experience, I’ve been using it to help me code my portfolio website and the results are amazing. I’ll share the whole process on that soon.

What’s making her not cool:

It can’t reference information from other chats because Claude doesn’t have memory of previous interactions. So, if I hit the limit in one chat, I have to start a new one and feed it appropriate info in order to continue my tasks. Even though I understand that it’s due to their emphasis on privacy and ethics, it’s definitely inconvenient at times.

Perplexity AI

Access Link

The lowdown:

Perplexity is an AI-powered search engine that provides real-time, sourced responses. It’s positioning itself to directly compete with Google Search, but also honing in on being a platform. Their launch of “Shopping” shows that they’re pushing towards changing consumer habits.

Quick facts: She’s from the Perplexity family, which are kind of in their own lane. She gets along better with Sophomores and Juniors.

Sit with her if:

• You want to enhance your activities like online shopping and research.

• You work in an industry where fact checking, breaking news, and finding verified sources is really important.

• You want a space to collaborate with others while using the power of GenAI.

• You like optimizing your everyday, and having one space to do multiple things instead of using 1000s of different platforms and apps to optimize your day and process

Why she’s cool:

I really like how Perplexity is positioning themselves into an information and productivity ecosystem and how they are optimizing specific flows that an average user of the internet experiences everyday. You have your main landing page which is a classic search bar, but below it you will find a few quick links that the system thinks you’ll be interested in.

Imagine opening your browser first thing in the morning and this is the screen that greets you. Amazing - especially because if you click on the “Discover” section, the next thing you can look at is the latest news, and with time, the latest news most relevant to you.

What I’m trying to get at is, the team at Perplexity is clearly targeting the everyday office employee. Every morning, the average person takes a seat at their desk, coffee in hand, powers up their computer, opens their browser to see the Google search bar and then first checks out some of the latest news. Then, they move over to their team channels to check out what’s been going on and determine their first task of the day. Seems that Perplexity thought - why not optimize this flow and instead of people opening 5+ different tabs and windows to get their day started, offer it all on one platform? I really look forward to seeing how they grow.

What’s making her not cool:

Haven’t used it enough to hate on anything in particular. Maybe I’m not really loving what they’re hiding behind the paywall? If you want mass adoption, you need to give people a reason to want to pay for your product, and I’m not sure that their free version is giving enough.

MidJourney

Access Link

The lowdown:

MidJourney is an image generation platform that uses AI to generate images based on your prompt. That’s really it.

Quick facts: Her parents are MidJourney and she’s friendly to Freshmen, but only if you have money.

Sit with her if:

• You want to experiment creating fun images

• You’re not graphic design savvy and need some help creating custom graphics

• You’re an artist or designer that wants some extra tools for brainstorming and inspiration

• You’re looking for an alternative to stock images

• You don’t mind paying for a subscription right off the bat

Why she’s cool:

I’m not big on using AI to generate images, especially as someone whose career is highly reliant on my skill to generate visuals, whether they’re complex UIs or a simple email hero. However, I like that the platform itself is easy to use. Very straightforward in terms of what you can do when you land on your dashboard - start exploring or jump right into creating images.

What’s making her not cool:

In order to even try using it, you need to subscribe to a plan. This is fine if you frequently use image generating features and are already familiar with what MidJourney can offer you in comparison to the rest. However, if you’re a first time user, why would you commit $10 right off the bat to try out the platform? Some type of first time user option should exist.

Galileo

Access Link

The lowdown:

Galileo is an AI-powered system that uses natural language prompts to generate wireframes, mockups, and UIs for mobile and desktop screens. Pretty niche, but I have an inkling a lot of you will actually find this useful.

Quick facts: She’s not realted to Galileo Galilei, but her family Galileo is like trying to revolutionize product design. She’s friendly to Freshman overall, but you’ll get along better if you’re a Junior or Senior.

Sit with her if:

• You need to conceptualize an app or platform and want some quick ideas

• You work on a UI/UX/Product Design team and want something effective and quick for rapid prototyping

• You have an idea for an app and want to visualize your idea

• You’re a designer that wants to test a few different designs quickly

• You’re a developer that has no designs to work off of

Why she’s cool:

I really like that this exists. Is it an unpopular opinion amongst my colleagues? Maybe. However, just like Canva didn’t replace graphic designers, Galileo won’t replace the need for UX Designers. What it does do, is help people who have a good idea, but lack the design software skills to execute it.

A platform specific feature I like is that it gives you an option to generate a UI from text or from an image. I personally love sketching new screens, and the Image to UI feature allows me to upload my hand drawn sketches and create a first pass of a UI that can be moved over to Figma for further iteration.

What’s making her uncool:

Something that I would love to see in the future is some type of personalization layer on the platform. For example, if a company like Uber is using this as a tool, Galileo could learn more about Uber’s design system, flows, and users and offer curated experiences based on the needs of a particular project.

DeepSeek R1

Access Link

The lowdown:

The new kid on the block. DeepSeek R1 is an AI assistant being labeled as a cheaper alternative to GPT-4 and superior at logical reasoning when compared to everything else that currently exists on the market.

Quick facts: No one was paying attention to her until she got a major glow up recently. She hangs out with Freshmen a lot, but now that she’s suddenly hot all the Juniors and Seniors are like obsessed with her.

Sit with her if:

• You’re a startup that needs an AI tool, but also needs to save money.

• You need help with coding and software development.

• You need advanced reasoning for complex tasks and research.

I haven’t had a chance to use it yet, so the outlined use cases are according to what others have shared and what I’ve found in my research. Once I am able to gain access, I will adjust this section.

Other AIs I’m not really familar with yet but seem relevant to list here:

Ideogram AI

The lowdown:

According to ChatGPT - text-to-image generation with enhanced typography capabilities.

Rumor has it:

Better text rendering in images than MidJourney.

Uizard

The lowdown:

AI-powered no-code design tool for prototyping apps and websites.

Rumor has it:

Converts hand-drawn sketches into wireframes.

Runway ML

The lowdown:

AI-powered video editing and generation platform.

Rumor has it:

Apparently has really food AI-generated video editing (VFX, motion tracking, object removal).

Command R

The lowdown:

Enterprise-focused AI with retrieval-augmented generation (RAG), designed for businesses that heavily rely on fact checking.

Rumor has it:

Offers enterprise document analysis and summarization and customer support automation.

So, who have you been sitting with so far and is there an AI that’s giving you FOMO now? Share your thoughts in the comments.

Leave a comment

Let me know if I missed any major players in the consumer GenAI space. Hoping to adapt this list as time goes, which I will be sure to indicate whenever I edit the list or add a new option.

Happy Monday, brats!!

I didn’t waste my time away, though. I was in the final stretch of a major Fortune 100 website redesign project I have been working on this past year. I also had the honor of talking to Cristian Pineda on his podcast. This was my first time being a guest on a podcast and it was a thrill of a lifetime talking all things IA, UX, product design, and AI. You can check out the episode below.

Cristian Talks Product Design

Adapting with Purpose

Welcome to Issue #16 of Cristian Talks Product Design…

Listen now

a year ago · Cristian Pineda and JELENA

This week’s roundup is a little different than what you’re used to. I will be experimenting with different categories, sections, and formats of this weekly roundup for the next few weeks, so I’m excited to see what everyone thinks. Share your thoughts, feedback, and ideas in the comments!

☆ Bookmark this

// Bloomberg’s AI Glossary

This Glossary for the AI Revolution is a great bookmark to have in your bookmarks bar. It outlines all the relevant AI terms that are being thrown out there in the news and in the corporate world daily.

⌨ Relevant to your 9-5

// AI is taking over customer research

Every has become sort of a comfort publication for me. Their articles just scratch an itch I have for certain types of discussions and topics that I didn’t even know I had. So in one of my recent reads, I stumbled on this article by Chris Silvestri diving deeper into his empathy engineering framework. When I saw the term “empathy engineering” a lightbulb went off in my head. I had to go back and read the other articles in this series to figure what the hell empathy engineering even is.

Chris introduced his framework in Decoding Your Customer With AI. Here’s how he lays it out:

❝Instead of just imagining what your ideal customer wants, you can actually have a conversation with an AI model that mimics their persona. You can prompt it with detailed information about your target customer—their demographics, goals, pain points, even their favorite industry websites. Then, you ask the AI questions, just as you would in an interview with a human customer. How would they describe their biggest challenge? What features are most important to them? What are their objections to your product?❞

While his framework is geared towards customer research for copywriting, it can absolutely be tweaked and adapted for any type of customer or user research. Is this what they meant when they said AI will take over our jobs? Looks like User Researchers might be in a pickle. Jokes aside, I’m curious to see how far we can go with effective AI prompting, and how accurate can these results really be.

⛱ Relevant to your free time

// AI Hacks for Black Friday

In the spirit of this week, I was hellbent on finding something juicy around shopping and Black Friday deals.

Forbes released these 5 AI Hacks for Black Friday shopping, while I took a look at Perplexity’s Buy with Pro feature and dove deeper into Walmart’s AI enhancements to shopping. Finally, the AI direction I’m looking for.

Personalization is what I’m craving when it comes to AI. These apps and platforms know so much about me, so tell me why am I still having to sift through dozens of websites to find a “cool 70s inspired funky but classic lamp” ?

On Friday I will be powering up a few different AI platforms and utilizing AI features on various eCommerce sites for research purposes. I want to see how helpful they really are, how well they are handling the Black Friday pressure, and most importantly how usable are they? Can’t wait to come back with my report next week.

☕ Interesting Opinion Pieces

// Who needs an AI doppelganger?

John Herrman of New York Magazine wrote about Meta’s AI doppelgangers and it got me thinking about what constitutes an AI as ridiculous? Who determines the line between giving the user what they don’t know they need and making up dumb ideas?

The AI doppelgangers Meta introduced at the beginning of the year are nothing short of impressive. However, it definitely goes into those sensitive AI waters that most people have been worried about. How do we make sure such technology doesn’t get into the wrong hands and how do we protect ourselves from having our likeness turned into an AI doppelganger without our consent?

On the other hand, the AI chat integration into Meta’s messages is definitely something I can see evolving into a very usable tool in the future. It seems that the current version is still very robotic, as expected. You can’t expect an automated bot that’s pretending to be you to emulate your style of writing, personality, and overall human characteristics. However, I’m excited to see how this feature evolves and how it will be integrated in our everyday workflows in a few years. Imagine Microsoft Teams implementing a feature like this? Maybe then we’ll be able to avoid those annoying “Hey” messages that don’t follow up with anything else.

⚡ News You Shouldn’t Miss

// Perplexity releases shopping assistant tools utilizing AI explainability

Just in time for holiday shopping and Black Friday, Perplexity launched “Buy with Pro” that offers product recommendations and “Snap to Shop” visual search that can help you find the product you are looking for. It’s opening new doors for companies to market their products and most likely going to be a key player in this new era of shopping and marketing.

// Amazon invests $4 billion in Anthropic

Amazon is the latest company to hit the gas pedal in the AI race by committing to investing antoher $4 billion in what’s being dubbed the “responsible” player in AI, Anthropic. What this means for the future of GenAI and how will all these GenAI companies start to differentiate themselves, only time will tell.

// Microsoft turns 50 and is more relevant than ever

As someone who knew how to install a video game on Windows 95 before even knowing how to write her name, Microsoft has always been an integral part of my life. However, I was too young to be aware that in the late 2000s, Microsoft wasn’t that cool anymore. This Wired article takes a look at Microsoft’s renaissance and really makes 50s look like the new 20s.

Love & Hate Mail

All types of tantrums welcome at hello@jelenacolak.com.

Current Learning Phase: Research and Information Absorption

Educational Resources

Last week I shared the UX of AI website as one of the educational resources in this section. What I like most about it is how the website itself outlines 16 design principles but then under each includes more links for further learning and research. In attempt to make this learning process a little more community oriented, I started a special study chat for everyone to join in each week.

Last week’s homework was to take a look at the first AIX principle and share what other resources we’ve all stumbled across in our readings and research related to the principle. I’m extending the first principle homework and discussion to this week as well, so please join in and share questions, thoughts, and useful links in the chat.

In the meantime here’s what I found:

Wizard of OZ User Research Method

In my readings for our first tech brat homework assignment, I stumbled upon the concept of the Wizard of OZ Usability Test. It’s a great way to test your ML system for possible gaps in design of the recommendation system. It is done as part of a moderated usability test where one team member facilitates the test, while the other acts as the system in the background, giving back results based on the test user’s interactions. I’m pretty sure I’ve done something of the sort in one of my first UX college classes.

Designing Machine Learning Systems

This is the first time I am featuring a book on the list, and a technical book might I add. Since I just bought it last week, there aren’t many insights I can add right now. However, I’m mentioning it here in case there is someone that might find interest in it as much as I did. In order to be able to design products, there is a certain degree to which you need to understand how they are built. This is the case for Machine Learning Systems as well. Anyone who tells you otherwise is giving you bad advice. This book seems to have the perfect blend of technical and strategic information that helps big picture thinking, and if you are thinking about designing some type of AI based product, what type of system you need to build. Looking forward to diving deeper into this book and share what I’ve learned.

Five UX Principles for Designing AI Models

It was refreshing to find a publication that focuses on the same POV I’m focusing on with my writing. writes where a lot of his content is bridging the gap between UX/UI Design and AI. This article takes a look at some familiar design principles and helps us adapt them for the new era of products.

Interesting Opinion Pieces

Research shows customers don’t trust AI

An interesting discussion popped up in the Reddit UX Design community the other day - the true effect of AI implementation not on just user experience but customer experience. Consequently, that thread led me to this link from Marketing Dive. I’m not surprised consumers are skeptical.

First, a lot of people do not understand the technology, yet businesses and people in tech that do completely disregard that. Consumers haven’t been introduced to AI properly, and if it’s a bit of a scary thing for us in tech how can it not be to an average Joe? Second, most AI experiences suck. I talk about that all the time.

Just this past Sunday I was on GoDaddy trying to get an agent to talk to me through their chat, and it took me a little bit of effort to bypass the annoying chat bot. Not to mention how many times a day Meta AI starts messaging me on Instagram to my dismay. I keep reiterating how our approach is all wrong - we keep trying to implement AI into existing workflows when really we should be reimagining workflows with AI as the system supporting them. I’m trying to think of ways to start spreading this message more strategically, stay tuned for whatever I come up with.

Make your life easier

New Notion AI can help visualize flows

The first time I’m seeing AI potentially taking away some aspects of my jobs is this example of Notion AI creating a diagram purely from the user’s existing content in the platform. It doesn’t worry me. On the contrary, it shows how there might be some unexplored ways to tap into this product and streamline current processes. When creating large products, there is miles and miles of information about the product written in requirement docs. If you were to stick those requirement docs into Notion, it could save not just designers time but also PMs. I can see it helping with gathering all complex requirements and creating a first pass design for certain flows, or maybe even the IA for an entire platform. Time will tell, but I degfinitely suggest logging onto Notion AI and checking it out!

News You Shouldn’t Miss

Last week felt like the most insane AI news cycle ever. Maybe it was because I was watching it very closely and in tune with the reactions on X. Either way, lots of interesting developments that have me on the edge of my seat.

OpenAI Developments

Monday: Sam Altman publishes his Intelligence Age essay and Tech Twitter’s reactions are mostly along the lines of “forget founder mode, bro is in god mode,” and not in a… inspiring way.

Tuesday: OpenAI announces rollouts of Advanced Voice. I have yet to check it out because I myself have been slow to accept ChatGPT and am still getting friendly with it. I also was never a big voice command girl, but of course I will put some time aside to try it out and report back.

Wednesday: OpenAIs CTO Mira Murati announces she is stepping down and a few hours later we hear that OpenAI is considering giving Sam Altman a 7% equity stake in the company and restructuring to become a for-profit business. My inkling is that this has something to do with that Jony Ive project. 🕵🏻‍♀️

In other news…

Meta gives us a glimpse of their Orion glasses and Zuck talks about how he can see glasses become what the smartphone became to a computer - not a complete replacement of technology, but a leader in devices we use in our everyday lives.

Notion launches its newly improved AI, which you were porbably aware of by now if you read the “Make your life easier” section of this letter. 😉

In another universe…

Not a lot from the other universes this week. Ya girl was too focused on AI topics to even notice if anything else is going on.

Got beef?

Feel free to send hate mail directly to hello@jelenacolak.com.

Current Learning Phase: Research and Information Absorption

Educational Resources

AI Terminology Cheat Sheet

I love this list of all the AI Terminologies we stumble upon when researching about this evolving technology and reading related news. I’m using it as a reference list for my learning journey and letting it guide me.

UX Patters for AI

An insanely great resource created by , Shape of AI is today what the Laws of UX website was for us designers 10 years ago. It serves as sort of a resource center on important UX patters for designing AI supported products and interactions. As someone who had formal University education in User Experience, I really hope professors are sharing this resource with their students and that these are slowly being integrated in UX curriculums so the next generation of designers can be ready for what’s coming.

UX of AI Guide

Another awesome resource I found which dives deeper in the process of designing an AI. I will be using this to guide my further research for weeks to come and will be going through my research weekly along with the process outlined here. Hoping we could maybe make this a collaborative learning process. I will share more details in the chat but ultimately would love to come back to the chat every week and see what everyone has found on their journey and share it with the group.

Interesting Opinion Pieces

The Button Problem of AI

”Every incremental innovation in technology starts off as a differentiated skill but ends up as a button.

Evan Armstrong writes about the current state of AI being basically just a lot of buttons prompting a system to do the work that you would have done yourself anyway. He gives examples of his workflow in the email app Superhuman and concludes that, while it helps his productivity, nothing in AI today really has a WOW factor. He concludes with this thought provoking statement:

We're so focused on making AI fit into our existing workflows that we've forgotten to ask whether those workflows even make sense anymore.”

You know how I keep reiterating that UX Design will become AIX Design? We are entering what might be the era we entered when Google was invented or when Apple introduced the iPhone. Some apps and software digitized existing analog workflows like OpenTable or QuickBooks, some reimagined an existing sector like Uber or Airbnb, and some introduced a completely new idea like YouTube or Snapchat. Our close future is reimagining existing flows or introducing new ones, with AI in mind.

“People are seriously underestimating App Intents”

Some developers are having more tech focused discussions around Apple Intelligence that I’m really excited to see. It points in the same direction I keep pointing in and actually touched on in one of my recent LinkedIn posts:

“Prediction: Apple Intelligence will spearhead a wave of new AI powered apps, leveraging what the new iPhones and iOS offer and incorporate it in the experience.”

In some way, the new and improved Siri and Apple Intelligence will start a development wave of new types of apps. They might not even be called apps by the time these roll out. I’m in the process of really digging deep into the App Intents and Apple Intelligence in order to better understand what’s all happening under the hood over there and how it can potentially be leveraged. This is definitely a topic you will see me write about a lot.

Make your life easier

Usually I like to feature a cool new AI powered feature, plug in, or platform that can potentially make your life easier. This week in the spirit of the rest of the content in this stack, I urge you to explore what Apple Intelligence and Siri can do to make your life easier. Read a little bit about it here.

News You Shouldn’t Miss

In case you live under a rock LinkedIn is training AI Models on your data, to opt out head over here.

The design world’s lord and savior Jony Ive is working on an AI Device startup with OpenAI.

The newest version of the device our lives depend on came out this weekend, but they aren’t fully equipped with Apple Intelligence just yet.

Next time you call the Polo Bar to try and snag a reso, you might not talk to a human.

In another universe…

You’ve probably seen the viral post circling around all social media platforms. Nike’s new CEO Elliott Hill work history:

From intern to CEO, Elliott Hill’s resume resembles that of a textbook example American Dream resume. Sure his loyalty was rewarded but it wouldn’t have been if he wasn’t the right person for the job(s) and a capable individual. What intrigues me is, especially as someone who works in the ever-changing Tech sector, would our generation be able to pull such a feat? Is loyalty indeed rewarded in this day and age? Is it possible to even be loyal?

Looking at the actual years Hill was at Nike, his career grew through iconic periods of Nike’s business, as well as the two decades of significant technological advancements and changes in consumer behavior. He was there while Jordan was still flying like an eagle and Air Jordans were leaving their mark in fashion and culture. He was there for the dot-com boom and the social media boom. He was an integral part of what drives revenue for this company during times of big shifts, making every year and every role interesting, dynamic, and I’m sure what almost felt like working for a different company.

My generation is not so quick to stick with the same company for years. There are immense growth and compensation benefits to changing companies every few years. Many also like a “change of pace” whether its different company cultures or different types of projects and sectors. Some people don’t aspire to climb the ladder to leadership and are happy being individual contributors for the rest of their career. I wonder if in 20 years a similar post will float around for a Millennial or Gen-Z CEO. We have yet to find out, but until then read a little about Nike’s changes.

Got beef?

Feel free to send hate mail directly to hello@jelenacolak.com.

Current Learning Phase: Research and Information Absorption

Educational Resources

Machine Learning Models

Remember how last week I touched the surface on the different types of AI Models? Well, one of the opinion pieces I share a little later in this newsletter prompted me to direct my research to Machine Learning Models specifically. Really, that’s what I have been mostly interested in general when it comes to this whole AI thing. If I were to just get a glimpse of the Netflix recommendation model, I would die a happy woman. However, I must satisfy myself with reading these somewhat complex articles. Here’s a good one by Hewlett Packard Enterprises. Now, this is getting a little more complex but I think it serves as a really good glossary and jumping off point for further research, as well as a good bookmark to have handy.

Interesting Opinion Pieces

How AI’s Booms and Busts Are a Distraction

An interesting piece by Vox’s Kelsey Piper on the current state of our approach to AI safety. It’s true - most people think about it from a perspective of “oh no, super robots are going to take over the world,” but there’s much more to it. Just look at how policies around social media have been developed in the past 10 years. We should learn from that and apply the experience to the developing world of AI.

Human Centered Machine Learning

My UX people will find this one particularly interesting. It’s an opinion piece from 2017 written by Jess Holbrook for Google Design. Jess works at Microsoft now, but what he’s written here is still relevant. Number 6 on the list definitely got me excited because it aligns with what I’m thinking might be the future of our roles, especially as someone that has been working on Information Architecture projects and has worked on setting up data labeling strategies. It’s interesting to see how much of this article is still relevant and how much the industry has evolved in a slightly different direction. What’s written here are the types of conversations we should be focusing on.

Make your life easier

AI Assisstant for Interface Design

I debated gatekeeping this one, even though I’m sure many of you know about it. Galileo AI is a UI generation platform for easy and fast design ideation. Their words, not mine. This is what I want to see implemented in Figma in some capacity. I also think this is a great tool for someone that might be ideating on a product but can’t afford to get a designer involved.

News You Shouldn’t Miss

Companies Finally Find a Use for Virtual Reality at Work

“Finally” is a word the WSJ is using. Remember two years ago when there was this weird wave of VR theorists convinced we will be taking meetings in a Second Life type of online environment? Yeah…

Using AI to Detect Breast Cancer

This past week I stumbled upon this image on my Instagram feed posted by The Female Quotient and it intrigued me.

This is what I want to see more of. This comment also reflects what I mean when I hate on the conversations around GenerativeAI - it’s always companies rushing to hop on the AI train and their first solution (because it’s the easiest) is introducing some GenAI assistant that literally no one is using. Anyway, after a further deep dive I discovered this is not new news but nonetheless it’s awesome news. Read more about it here.

Forbes releases its “Next Billion-Dollar Startups” List

Read the full list of 25 venture-backed startups here.

Got beef?

Feel free to send hate mail directly to hello@jelenacolak.com.

I feel disconnected from a lot of newsletters, especially business focused or tech newsletters. They feel very academic, very “official.” I mean, I guess that makes sense considering they’re written by serious people, or so it seems. I guess I don’t consider myself so serious even though I’m a still-very-young-year old Associate Director of Information Architecture and User Experience.

Realistically, calling a newsletter about what are considered more serious topics “tech brat” makes me unserious. I don’t mind it though, because I’m tired of the usual positioning of these topics. Yes, there should be tough and serious conversations around AI but I also believe that in order to spearhead adoption and help peers adapt to the changing landscape of technology, we should make learning about AI more accessible and more fun.

So, welcome to tech brat, a weekly newsletter rounding up resources about AI and related tech topics meant for designers and other non-technical tech creators, operators, and leaders.

Let the record show that I named this tech brat before Charli XCX came out with her LP “brat.” The unfortunate reality is that I waited over 6 months to start writing, hence why I went back and forth deciding if I should keep the name. I felt connected to it, so I’m keeping it.

What makes this newsletter bratty is that it’s written by me, but also, it will probably evolve from a “for AI newbies” newsletter to something else, as my knowledge of the space evolves. You will also notice that I take some pretty bratty stances around AI, like for example, my disdain of conversations around Generative AI.

I believe having a beginner sharing resources makes this publication stand out from others that have experts talking to beginners. Nothing wrong with experts trying to share ideas, knowledge, and guidance, there’s major benefit in such publications. However, I think there’s something more approachable and more fun about learning together.

So here’s the first of hopefully many weekly roundups that will include links I’ve found helpful or interesting throughout the past week, and the occasional opinion section or wild piece, sent directly to your inbox.

Current Learning Phase: Research and Information Absorption

Educational Resources

The recipe for building AI Models

Recently I started seriously reading about AI in relation to How AIs are built. I wanted to read more about what the current process is like because I realized that through conceptualizing new products and working on the initial experiences and Information Architecture (IA) concepts for their MVPs, I myself have lead the ideation of a few AIs.

The first article that really confirmed I have gone as far as identifying what type of Algorithm might be appropriate for a certain AI was this LinkedIn article by Daniel L. Building Your First AI Model: A Beginner's Step-by-Step Guide

This had me wondering how the role of UX Designers might evolve. Consequently, it also had me wondering if my involvement in these types of projects was more by accident. Is it due to my Information Architecture background? Is it due to me just being the go-to person to bring on when conceptualizing products and digital solutions and not an indication of where the industry is moving? Will positions such as “AI Architects/Designers” be a thing in the near future? Something to think about.

That topic aside, here is another valuable link on Building AI Models:

How to Create Your Own AI Model from Scratch? - This one breaks down the steps a little more and also starts including some code snippets. I personally have some experience with Python and some other coding languages so the snippets actually give me a little bit of context to “visualize” what might be going on in the background. However, even if you don’t it’s a valuable resource to kick off further research.

The main types of AI Models

After the epiphany that any good strategic thinker with a semblance of creative problem solving talent can work on designing an AI Model (controversial opinion?) I started looking at different types of AI Models.

This random blog post that’s intended to drive traffic to this company’s enterprise solution was a surprisingly good jumping off point to start my research around the different types of AI Models. Now it’s time to really do my homework and provide some, maybe actual academic resources(?) for next week.

I would like to see where these links might lead you in your research, so please share what you find helpful in the comments or chat. The point of tech brat is to create a community of newbies that share valuable info with each other, so let’s get cookin’!

Interesting Opinion Pieces

The rise of the Model Designer

In my search for any indication of what the industry thinks about UX moving into “AI Model Design” I have found a lot of articles about how UX Designers should be utilizing AI in their work, and not many on how they should be learning about designing AIs. However, this gem of an opinion piece from Paz Perez for UX Collective on Medium, makes me think that I just might be ahead of the game (along with her) when thinking about the future of our work.

GenAI Will Change How We Design Jobs

Even though I have a certain aversion towards many Generative AI topics, Paz Perez linked this Harvard Business Review article as a source in her opinion piece. It’s a valuable read on the general state of how GenAI will change how we design jobs.

How Do You Change a Chatbot’s Mind?

Now this is the type of content I want to see about GenAI. NYT’s Kevin Roose wrote about changing ChatGPT’s, Llama 3’s, and other GenAI’s opinion of him. An interesting quote popped up here as well:

“Profound does what it calls “A.I.O.” — A.I. optimization — on behalf of Fortune 500 companies and other large businesses. (…) Mr. Cadwallader and Mr. Babbs believe that A.I.O. is the natural successor to S.E.O. — search engine optimization”

You best believe this quote alone will keep me busy this week.

Make your life easier

AI Powered Budgeting Tools

I’ve been out of school for years now but I can’t escape the back-to-school feeling I get every September. I was always a fan of Fall. September 1st to me is what January 1st is to most people. I stumbled upon this Substack Note by outlining pros and cons of AI Budgeting Tools and I restacked it immediately to have it handy for when I’m ready to reevaluate my monthly budget, which is definitely on my list sometime this month. If you check out any of these or already use them, share your thoughts!

News You Shouldn’t Miss

OpenAI agrees AI generated content needs designer label treatement

Investors are shorting AI companies

Android Gmail users are one step closer to the 21st century

Voldermort of the design world is making a lot of social media managers really mad

Are you smarter than an A.I.?

Here’s a fun quiz to test your eye sight and how well your “Where’s Waldo?” skills have improved since grade school - Quiz: Fake or Real Images?

In another universe…

This past August there have been many shakeups that I personally thought were interesting for the future of business and possibly design.

Starbucks hires a remote CEO

An interesting topic that popped up on my LinkedIn the week the newest Starbucks news dropped was “How to Fix the Starbucks Experience”.

Starbucks has an opportunity to adapt their customer experience to changing times. Their customers rely heavily on ordering through the app because the in-person ordering times are unreliable. However, the app has also been lacking, contributing to overall customer dissatisfaction. Instead of brainstorming what the next pumpkin spice latte concoction should be and how it could be marketed, why not bring some AI innovation to the app and extend it to their in-store experience?

Got beef?

Feel free to send hate mail directly to hello@jelenacolak.com.