AI Isn't Why 95% of AI Projects Fail.

When I was learning to build with AI, I realized we're hiring wrong.

As a solo founder, I could wear every hat. Idea, system, verification, direction, craft, all of it, because AI can execute so much of what used to require a specialist. I didn't need to be a great coder to get working software. I didn't need to be a trained designer to get something that looked right. AI closed that gap fast.

But here's what I noticed almost immediately: each hat is still its own kind of thinking. AI collapsed the skill gap between them. It didn't collapse the thinking gap. Writing code and deciding what's worth building are not the same judgment, even if AI can now help with both. Making something look good and knowing it feels right to the person using it are not the same judgment either. AI can execute all five. It can't decide which one needs your attention right now, or catch it when one is quietly wrong.

MIT put out a stat that stuck with me: 95% of generative AI pilots at companies are failing. I don't think that's an AI problem. I think it's a hiring problem. Companies are still hiring for job titles built for a world where execution was the scarce thing: the coder, the designer, the strategist. AI just made execution cheap across the board. What's left, the actual scarce thing now, is judgment. And most companies haven't rebuilt their hiring around that at all.

We Need to Rethink What a Coder Is. And a Designer. And Everyone Else.

AI can write code that runs. AI can design something that looks clean. AI can draft a roadmap that sounds reasonable. It does all of this at a level that would have taken a specialist years to reach, and it does it in seconds.

So if you're hiring "a coder" hoping they'll write good code, you're hiring for the 95% AI already does. The part that actually matters is the 5% AI can't touch: knowing whether the code is solving the right problem, catching the assumption that'll break in production, knowing when to stop building and ship, knowing if it actually feels right to the person using it.

That 5% isn't one skill. It's five different kinds of judgment, and almost nobody is strong in all five. What you're actually looking for, when you hire now, isn't a coder or a designer. It's a polymath: someone who can move between these modes of thinking, or at minimum, recognize which one a problem needs and know when they're not the right person for it.

The Five Kinds of Judgment AI Can't Replace

Idea. Knowing if something's actually worth building, not because a spreadsheet says so, but because you understand the person who'll use it deeply enough to feel whether it lands. AI will hand you fifty ideas before lunch. It can't tell you which one is real, because it has no stake in any of them. Weak here, and everything downstream is noise wearing a nice interface.

System. Understanding how the parts connect. What happens when you pull this thread. Where the whole thing snaps if you're not careful. AI will build you any single piece beautifully. It won't tell you that piece contradicts the one you built yesterday. Weak here, and you're building a house on sand.

Verification. Finding what everyone else missed. Looking at what's been built and knowing, immediately, here's where this fails, here's what we're assuming that isn't true. AI can check its own work against the instructions you gave it. It can't check those instructions against reality. Weak here, and you ship broken and pay for it in public.

Direction. Knowing what comes next, and just as importantly, what doesn't. Saying no to the feature that doesn't serve the real thing, no to the expansion that dilutes it. AI will happily build whatever you ask for next. It has no opinion on whether you should be asking. Weak here, and you build in fifteen directions and finish none of them.

Craft. Making it actually feel right. The pixel that's a hair off. The interaction that feels calm instead of anxious. AI can generate something polished. It can't feel whether it's actually calm to use, because it's never had to live inside the thing it made. Weak here, and you can have the smartest product in the world and nobody will want to open it twice.

Here's the hard part:

You can't just ask someone if they have these. Anyone can say "yeah, I'm good at strategy" or "yeah, I catch edge cases." Self-reported skill means almost nothing here, because these aren't credentials, they're instincts, and most people describing them have never actually been tested against all five.

The closest thing to a real signal isn't experience or a portfolio. It's curiosity. The people who end up strong in one of these, and genuinely aware of the other four, are the ones who are inquisitive by nature. Who ask "why" one more time than is comfortable. Who adapt instead of defending the way they've always done it. That restlessness is what builds the judgment in the first place, and it's the only thing I've found that reliably predicts it.

Hire for that. Not the title, not the tool, not the confident answer in the interview. The curiosity that got someone there in the first place.

I built Tansei while learning AI, and wearing every hat myself is what taught me this. It shipped July 19th. Free for 3 days, then pay once.

→ 🔗 https://tansei.io/