I did product engineering before it was cool. No-code was the reason why.
Bubble (the major no code platform) was full of constraints. If you wanted to build something custom, you could not just write a clever script to hide a flaw. You had to work within its boundaries. Because it was a visual way of building, I did not have to worry about syntax and started obsessing over user problems. I spent less time hunting down typos and more time asking why a user needs this tool anyways.
That shift changed everything for me. Implementation became secondary, the product became the entire focus.
Today, building software is even faster. Harnesses like Claude and Codex can generate an entire application in minutes. But as building gets easier, we are making a dangerous mistake: we are forgetting how to think for ourselves.
The limits of a tool teach you how to think
When you strip away traditional programming, you are left with pure logic. Bubble co-founder Emmanuel Straschnov noted that visual logic forces creators to map out user workflows action by action rather than worrying about raw syntax. You have to understand state, data relationships, and the steps a real person takes to get value from your app.
That is what product engineering is all about. As the team at PostHog wrote, a product engineer cares more about the outcome and impact on users than the implementation. Gergely Orosz made the same point in The Pragmatic Engineer: the strongest builders want to know why product decisions are made, challenge specifications, and focus on how people actually use what they build.
When the tool takes care of the plumbing, your only job is to understand the human on the other side of the screen.
AI is a generalist, but you are the specialist
A lot of founders now believe they can outsource the thinking to an LLM. They prompt a model, take its suggestions as gospel, and assume the machine knows what users need.
It does not.
AI is a generalist trained on past text. It knows averages, patterns, and common denominator answers. It does not know your customer. It does not know the specific friction that makes someone close a tab in frustration. You are the specialist. You understand the pain, the context, and the nuance of what people actually want.
There is also a massive gap between the tech world and reality. Inside our bubble, it feels like everyone lives in an AI chat interface. The numbers show a very different picture. A 2026 study from the Reuters Institute and the University of Oxford found that only 8% of Americans use AI chatbots daily. In the UK and France, that number barely reaches 3%. Most people who have tried AI have only used it once or twice.
Cognitive offloading ruins product judgment
When you stop working through hard product problems, your judgment gets dull. Researchers call this cognitive offloading.
In a controlled study by Anthropic researchers, developers who learned a new framework using an AI assistant scored significantly lower on comprehension tests than those who worked through the material themselves. They finished slightly faster, but they did not actually understand the system they had just constructed.
The same thing happens to product strategy. If you let Claude or Codex decide your features, your user flows, and your positioning, you end up with AI slop. It looks functional on the surface, but it has no soul, no real perspective, and no true utility.
At Voicepen, our entire foundation rests on this idea. AI content should never be generic filler. Whether you are turning a voice memo into a blog post or converting a podcast into written ideas, the output must sound like you. The AI can handle the transcription and the formatting, but the perspective has to come from a human voice.
How to build without losing your edge
You do not need to spend your days writing boilerplate code, but you must keep full ownership of the product logic. Here is how I approach building today:
- Define the problem before touching a prompt: Write down the user problem in plain, simple sentences. Talk through it out loud if that helps you clarify your thoughts. If you cannot explain the problem without technical jargon, an AI will not solve it for you.
- Treat AI as an assistant, not an architect: Technologist Andrej Karpathy advises builders to keep tight feedback loops and verify outputs rather than delegating full autonomy to models. Give the model narrow, clear tasks. Reject any solution that adds unnecessary complexity.
- Talk to users who do not care about technology: Remember that three out of four people do not use AI every day. Watch them interact with your software. Their confusion will show you real product flaws that no LLM can predict.
- Focus on outcomes over features: Every line of output should solve a distinct friction point. If an AI generates a feature that does not directly help the user reach their goal, delete it.
Conclusion
I will never go back to spending hours debugging manual syntax. Modern tools and AI models make building too fast and too accessible to ignore.
Speed of execution, however, is not product sense. The tools can write the script, but they cannot tell you what matters. I will gladly use AI to build, but I will never let a model decide what is worth building.