The AI-maxxing obsession is making founders worse

I keep seeing the same thing across early-stage teams. Founders who should be talking to users are instead running long AI sessions about whether their idea is good.

They come out of those sessions more confident.

That is the problem.

The yes machine

AI models are trained on human feedback. People rate helpful, encouraging responses higher than blunt ones. The result is a tool that is structurally bad at telling you your idea is wrong.

You can prompt it to be critical. Ask it to steelman the counterarguments, poke holes, play devil’s advocate. It will, and then soften the landing.

“While there are challenges, the core concept has real potential.” - has shown up in more AI responses than I can count.

It means nothing.

Founders who already believe in their idea use AI to confirm what they already think, then mistake that confirmation for research.

What actually goes missing

Real conversations.

Sitting with a potential user and watching where they hesitate — none of that is in the training data.

AI can simulate the interview. It cannot replicate the signal.

Actual market data. Real churn rates, sales cycle lengths, distribution costs — AI either does not have them or invents plausible-sounding versions you cannot verify.

AI will never give you hard no.

Nobody is in the room when you talk to AI. Nobody to look you in the eye and say they would not pay for this or this business idea is not viable/profitable in the long-term.

The end-to-end AI founder

I keep running into early-stage founders who believe they can ship a full product using AI alone. Product strategy, UI/UX, frontend, backend, marketing copy. The whole thing.

The product looks complete from a distance. Get closer and the UI is generic in a way designers recognize in seconds.

The information architecture follows the same pattern every other AI-generated app follows. The copy says “streamlined” and “powerful” without telling you what the thing actually does.

Any decent designer or developer can smell it. Users can too, even if they cannot name why.

The issue is not that AI produces bad work. It produces the statistically average version of what a product in that category looks like.

Average is not a product. It is a draft.

The social media layer

On top of this, there is a whole ecosystem making things worse.

X, TikTok, Threads and Instagram are full of accounts telling founders that AI is replacing designers, developers, and strategists. If you are not on the newest model, you are behind.

If you are not running 10s or 100s of agents in parallel on your Mac Mini, you are inefficient.

Some of these accounts are true believers. Many are not.

The big AI companies have large influencer budgets. High engagement drives algorithm reach.

“Okay, this is actually insane. I built the entire B2B layer using Opus 4.7 in a weekend” performs well regardless of whether the product is real, scalable, or used by anyone outside the screenshot.

The founder scrolling those posts at midnight is getting a distorted picture of what is actually possible.

What is underneath it

Running agents on the cloud, automating everything, shipping without doing interviews — there is a fantasy in there about building a business without the uncomfortable parts.

Not because these founders are lazy. Most are working constantly.

But the specific work of talking to people who do not validate you, sitting with ambiguity, learning a market slowly — it is uncomfortable in a way that opening another Claude session is not.

AI-maxxing produces outputs.

Next.js web apps, Figma designs, architecture diagrams, product timelines, wireframes.

Meanwhile, real discovery work often produces nothing for weeks except a clearer sense of what you should have built instead.

I do not think the obsession is primarily about efficiency. A lot of it is avoidance.

What actually helps

The things that help are boring.

Talk to people who have built what you are trying to build. Not AI summaries of their thinking — find the forum threads, the long podcast episodes, the interviews where founders describe what actually happened. The details that did not make it into the highlight reel.

Talk to your market directly. Do it badly at first. An awkward conversation where someone says they would not use your product is worth more than ten AI research sessions.

Use AI for repetitive and predictable work where you can verify the output. First drafts of documentation, boilerplate code, test case generation when you already know the right behavior. These work because you have enough context to catch the mistakes.

Do not use it to define your positioning, your product direction, your design language, or your go-to-market thinking. Those need to come from real understanding of your users.

AI gives you the average version of what someone in your niche does.

You do not want to be average.

The real answer

Stop asking AI what to do next and go talk to someone real.

Call a founder who has been at it for ten years and ask if you can buy them coffee. Read what practitioners in your space are actually complaining about in forums. Find one potential customer and ask to watch them do the work you are trying to replace.

Or do not. The universe will force those conversations eventually. It just charges more for them.

---

This article was inspired by Mo Bitar’s YouTube video. If you have not seen it, go watch it now.