- Tech Tales & Tactics
- Posts
- Wait...We Don’t Design Features—We Design Behavior
Wait...We Don’t Design Features—We Design Behavior
Product work used to mean features. Now it means behavior.
Sebastian Muehl
April 17, 2025
Product roadmaps were built for features. But AI isn’t a feature — it’s a system.
I’ve been thinking a lot about how AI gets added to roadmaps. First, it feels like something you can scope. Like you’ve always done it. Just like any other feature.
But once you zoom out, you realize it’s not just a feature you’re adding — it’s an intelligent system that changes how your entire product works.
And the users? They won’t ask for it. But we need to shape the product in a way that once users experience it bundled with AI, they never want to go back.
It’s time to define outcomes, not outputs.
So I started digging: which companies are getting this right, and how should we rethink product strategy in an AI-native world?
💡Product Features Turn Into Systems
1. Features as Differentiators
In the past, product features - one by one were delivering value. Based on user research, observation and user persona definition, a Product Manager would learn which exact feature and user flow would deliver value to what user.
PMs debate and score different features on criteria like impact (RICE Scoring). And then prioritize feature by feature.
2. The Shift: AI Is Not a Feature - It’s a System
With generative AI being added to products, it adds a whole range of functionality. It doesn’t just add one specific feature. It adds a system with which the users need to interact to get way more than just one use case solved.
Example:
GitHub Copilot - it’s not just a list of features, it’s a co-worker.
Notion’s AI is not just a writing assistant; it changes the workflow… it formats, transforms, and brainstorms with you.
Figma AI is your creative assistant that designs, cleans up, and prototypes for you.
My team keeps coming back to this idea of “co-creation.” Rather than siloed tools spitting out context-free outputs, we need AI that intelligently anticipates our needs and adapts in a complementary way. The key is that AI is only going to be as helpful as its awareness of what you’re actually doing day-to-day. Right now we compensate by prompting users for more context, but we need AI that proactively knows our specific goals, our established design systems, past work, and project contexts.
3. From Features to Behavior: New Product Strategy Questions
Above are examples that go beyond “we added a chatbot.” And the examples of GitHub Copilot, Notion AI, and Figma show that AI is not a feature update - they are behavioral shifts.
They change how users work altogether. And in turn, it must change how we think about building digital products. Hear me out…
The product strategy discussion changes and starts more like:
Should our AI act as an assistant — offering help? (something like text suggestions in Gmail)
Or should it be an agent that completes the full workflow based on text input? (agentic AI).
Once decided, this shift brings another level of complexity for Product Managers and Product Marketing Managers. You’re no longer designing a single feature or flow. You’re designing for a system — which behaves differently depending on many factors like context, input and the user.
Let me outline a few questions that popped into my head when thinking this through:
How do you make sure your user discovers the value?
How can you ensure the users onboard the way you think they should? And how do they discover the interaction with AI as you expect it? Or what if, based on user input, the AI steers in an unexpected and confusing direction?
What does “working as expected” even mean now?
We can’t possibly predict or foresee every output. So, how do your acceptance criteria change? What range of scenarios and outputs must your product handle gracefully?
How do you test an evolving system?
Traditional test cases won’t be able to test for “does the product keep its promise?” Can you even prepare for all the interaction paths and not just the happy path? Or are there many happy paths?
Which opens up a longer tail of questions:
How do you find and define edge cases with infinite output options?
How do you track user behavior across unexpected paths?
And most importantly, what happens if a user can’t find the path to their desired solution?
4. AI Stack = Your Product Strategy Now
Here’s the part we as product teams need to realize: adding AI isn’t just a UI change — it’s an architectural decision. Your product’s intelligence will only be as good as the system on which it is built.
Deciding on the system and AI Stack should not only be an engineering choice! This is a product strategy—a new kind of product strategy decision.
The questions PMs need to answer now:
What is our AI Stack?
What models do we use?
Will we (and how will we) fine-tune them? Or is out-of-the-box good enough?
Where is the data coming from, and how do we ensure clean, consistent data inputs?
How do we improve the model?
How will we be able to collect user behavior data from interactions with AI?
How can we minimize or mitigate hallucination risk?
How can we ensure the cost structure doesn’t blow up? (a classic with AI)
We need to track product performance based on these decisions. New key metrics will be: speed, accuracy, costs (since they are variable based on AI usage), and brand new: hallucination risk.
5. The Best Feedback Loop Will Win
What’s old is new again: The best product leaders focus on and optimize the hell out of their product feedback loop, not just creating longer and longer feature lists.
The loop will include
Capturing user behavior within AI interactions
Understanding the signals from that data to improve UX, model performance, or behavior
Ultimately, closing the loop between what the user needs and what the AI “says”
The PM work in this era is shifting.
As we define products, we are also shaping the intelligent system behind them — and teaching them how to behave.
What’s your take? Have you had to deal with any of these decisions yet? Please hit reply and let me know.
Have a great rest of the week,
How did you like this edition? |
Reply