You Can't Critique AI You Haven't Built With

On friction, product thinking, and why building beats consuming.

Feb 19, 2026

This week I got asked two questions.

First: Are you an AI optimist or a pessimist?
And second: Why are you looking at AI from a product builder perspective instead of just a technical one?

My short answer to the first question: It depends.

My short answer to the second question: Because understanding different AI systems and design philosophies contributes to critical AI literacy. That's the whole answer. Everything else is elaboration.

But the elaboration matters. It changes how you see every AI tool you touch. And it's the reason I started Product with Attitude.

Hey, I’m Karo 🤗
I’m an AI product manager, builder and thinker who spends most of her time inside AI systems to understand what they’re built to do. I write Product with Attitude to help people stop consuming AI passively and start building with it deliberately.

If you’re new here, welcome!
Here’s what you might have missed:
2025’s Most Absurd Product Decisions
I Refuse To Choose Between Ambition And Presence with Casey Hemingway
I Built You a Valentine’s App in 33 Minutes (Source Code Inside)

Subscribe 🤗

Things Are Bad (But We Can Build Our Way Out)

From the human cognition perspective, things are bad. And I assume, they will get worse.

A 2025 MIT study found that participants who relied exclusively on AI for writing showed weaker brain connectivity and lower memory retention; effects that lingered even after they stopped using AI.

We could blame everything that’s wrong with our cognition on AI, but AI is just the latest in a long history of automating away the parts that made us sharper. From GPS, autocomplete, spell-check and phone numbers memorized by our mobiles, we’ve been choosing convenience over friction for years.

What comes next is worse. With GPS, we offloaded navigation. With AI, we're offloading judgment. And unlike spell-check or GPS, there's no obvious ceiling on what AI can absorb.

Friction is where learning happens. And we’ve been engineering it out.

Build To Learn

Some people instinctively push back against AI. Others just let it wash over them, accepting the outputs without questioning what produced them. I get both impulses.

But neither one builds the muscle we need.

It's immersion - hands-on, deliberate immersion - that helps us understand why, and how these systems influence our thinking.

That's why I keep saying: build with AI. I don't necessarily mean launch a startup. I mean use AI for more than just a chat box.

Design your own workflows, assemble your own systems, put together micro apps that only you’ll use, test different scenarios.

For example:

Build a personal knowledge base that an AI can search, then compare its answers against a model without your context. The difference shows you exactly how much of an AI’s usefulness comes from your decisions versus raw capability.
Run the same image generation prompt across DALL-E, Midjourney, and an open-source model. The aesthetic defaults are wildly different, and those defaults reflect cultural assumptions their training data baked in.
Build a simple automation that routes different tasks to different models. Use one model for writing, another one for code. The act of deciding which model for which job forces you to understand what each one does well, and what it quietly gets wrong.

Accept the friction that comes with it. Build to learn.

Build To Verify

We all know that AI sounds confident even when it’s wrong. A hallucinated answer and an accurate one arrive in the same calm, well-punctuated tone.

With code, correctness is observable.
Building creates your own verification layer.

Try this: give an AI data you already know well. Say, your company’s last quarter.

First, the passive consumer way: paste the spreadsheet into a chat and type Analyze this. You’ll get a polished summary that sounds right. Maybe it is. You’d have to check every line to know.
Then, the builder’s way: collaborate with AI to write a skill.md file or a short script that tells the AI exactly what to look for:
- which metrics matter
- what good analysis looks like
- what to flag.

Now the AI works within your framework, and deviations from reality become obvious.

Same data. Same model. Completely different level of trust in the output.

Do this a few times, and you develop an instinct for the patterns: where models cut corners, where they over-generalize, and - most importantly - how to prevent it.

Consumers trust the output. Builders learn to test it.

Tools Are Never Neutral

Tools are never neutral. They carry the fingerprints of their creators.

You can’t separate a tool from the company that funds it.

Values travel through code.

Some models optimize for speed, some for the ‘‘wow’’ effect, others for safety. Those trade-offs are more than just technical details - they’re value statements.

AI outputs are shaped by alignment decisions we didn’t participate in. Critical AI literacy means reclaiming that context and learning to see the decisions that were made before we ever opened the app.

Once we see the values embedded in tools, we stop reacting to what they show us and start noticing what they don't.

Digital literacy, in the deepest sense, is pattern recognition.

Critical AI Literacy Begins Where Brand Loyalty Ends

When we use AI daily, we stop asking Which one is best? and start asking Best for what? That shift is not small. It’s the difference between being a consumer and being a practitioner.

But most people never make that shift. They pick one tool and settle in. They build habits around its defaults, absorb its tone, accept its boundaries as universal limits.

Over time, the tool’s way of thinking starts to feel like their way of thinking. The model’s alignment becomes their alignment; and they forget that behind each AI response, there’s a business model.

This is why I encourage everyone to use multiple AI systems, as a discipline. Each one exposes blind spots the others protect. Navigating those differences forces decisions that no single tool demands of you.

That’s friction. And we’ve established what friction does.

Abundance Gives Us Power, But Only If We Choose Deliberately

We don’t have to live inside the dominant ecosystem. We can assemble our own.

Abundance of tool choices shifts power from vendors to users, but only if users are willing to choose deliberately.

That “if” carries enormous weight - the power is theoretically there, but most people never claim it.

They default to what's popular, what's bundled, what requires the least evaluation.

Deliberate choice means deciding which model handles which task and why. It means noticing when a default setting serves the vendor more than it serves us.

At times, it means building our own stack to practice the kind of thinking that pre-built tools are designed to eliminate.

Cognitive strength is built the same way as muscle: resistance first, results later.

Building is Participation. Participation is Prevention.

The danger isn’t AI alone. It’s disengaged communities.

When we stay silent, systems get designed about us, not with us. The people who show up early - who test, who critique, who build alternatives - are the ones who can influence defaults. And defaults shape everyone else.

This isn’t theoretical.
Every platform you use today was formed during a window when a small number of active participants set the norms that billions now live inside. Facebook’s News Feed, Google’s search ranking, Twitter’s engagement metrics, all designed during periods when most future users hadn’t arrived yet.

AI is in that window right now. The models, the interfaces, the policies, the defaults about what’s helpful and what’s harmful, all of it is being shaped in real time by whoever shows up.

If that’s only corporations, the systems will reflect their priorities. If it includes engaged, critical, building communities, the defaults have a chance to shift.

The problem is collective, the answer isn’t. It’s personal. Pick a tool you've never used. Build something small with it this week. That's it.

The good news is that we don’t need permission to participate. We need a willingness to be early, to learn, and to share what we’ve learned. The tools will evolve. Our edge depends on evolving with them.

The Real Stakes: Intelligence By Design, Not By Accident

Here’s what I think about more than anything else in this space.

Bigger models aren’t automatically better futures.

An unconstrained AI maximizes output. A well-designed one maximizes human growth.

The risk isn’t that humans can’t think, it’s that we won’t need to. If nothing pushes our reasoning, our reasoning stops pushing back. The most advanced era in history could also be the most cognitively complacent.

Human intelligence shouldn’t survive by accident. It should be engineered for.

And this is exactly where product thinking becomes essential. Product thinking asks: Who is this built for? What behavior does it encourage? What does the user lose if the design succeeds?

These are the questions that determine whether AI systems amplify human capability or replace it.

Why The Product Lens Matters

Every section of this essay has been a product question, whether it looked like one or not.

Which tools carry which values? That’s product thinking.
Why do defaults capture users? That’s product thinking.
How do you choose deliberately in an age of abundance? That’s product thinking.

The product lens isn’t one perspective among many - it’s the one perspective that connects all the others.

That’s why it matters.

Critical AI literacy doesn't begin with a course. It begins the moment you stop asking which AI is best and start asking what it was built to do, for whom, and at whose expense.

Share this post with 3 friends or colleagues and unlock a 1‑month premium membership 🤗

What’s Being Built in the PwA Community Right Now

Theory’s over. Here’s what’s actually happening in our community right now:

Karen Spinner is launching CarouselBot tomorrow, and I'll be posting an interview with her where she takes us behind the scenes of the build.
Ileana is working on a new tech art project and she needs your help! It's one anonymous question. That's it.
Pawel Jozefiak built Wiz, a personal agent that, among many other tasks, helps him publish articles on a consistent schedule.
Mia Kiraki 🎭 built an AI Tutor for Claude, that won’t let you fake understanding.
Jeremy Wright - Marketer built an app for long-distance couples to help them stay connected.
Dheeraj Sharma shipped Markdown-to-Branded PDF generator that turns any markdown file into a professionally branded PDF in seconds. Let him know if you’d like to try it! Dheeraj just joined StackShelf; you can see his profile and freebies here.
Marcela Distefano built a guide about AI risk before deployment.

Share your project with us!

Community Reads

Single Page Sandbox: Localhost is the new open source by Nirav Bhatt

Join 12K Members and unlock everything you need to experiment and build with AI. From learning paths to prompt packs and code blocks, discounts and the community that makes it so special.