Software has never been cheaper to build. A handful of engineers now ship what used to take a team of fifty. So the bet of the moment looks obvious: if building is cheap, build your way to AI transformation.

Kevin Mannion named this moment in a post that went viral this week: we are at the peak of Mount Stupid, the peak of confidence on the Dunning-Kruger curve. Maximum confidence and minimum competence, at the same time.

Here is the thesis of this post in one line. The downslope of that curve (the resistance nobody surfaced, the edge cases, the long last mile where adoption dies) is real and predictable, and in most companies nobody owns it. The function that owns the downslope before you build is a discovery with a Go/No-Go gate. That is exactly what we do.

The bet is to build. The data disagrees.

The peak is seductive because the first half is true. Building really is cheaper. What breaks is the leap that comes next: if building is cheap, then building is the answer.

MIT NANDA’s report, The GenAI Divide: State of AI in Business 2025, measured it without mercy: 95% of enterprise AI pilots produce no measurable P&L impact, despite $30-40B invested. Not because the build failed. Because the organization was not ready for what got built.

And it carries a number almost nobody quotes, the one that should actually slow you down before you sign off on a budget. Buying from a specialized vendor or forming a partnership works about 67% of the time. Internal builds succeed roughly one-third as often.

Read that again. The bet Mannion describes, the build-your-way bet, is the path with the lowest odds of success, by the numbers. That is not a consultant’s opinion. It is what happened across 300 real deployments.

Nobody owns the downslope

So what is that downslope, exactly? Mannion describes it better than anyone: the resistance nobody surfaced in the workshops, the edge cases that were not in the requirements, the long last mile where adoption dies.

None of it is new, and that is the uncomfortable part. More than a million people are paid, literally, to see this coming: transformation leads, organizational development consultants, comms specialists, adoption managers, people and culture teams. They have been through this curve before, with ERP, with CRM, with digital transformation.

But, as Mannion puts it, they are not at the table. He cites that 52% of organizations make AI decisions without HR in the room. The CIO owns the budget. The forward-deployed engineers own the build. And nobody owns the downslope.

That is the hole. It is not a talent problem or a budget problem. It is that accountability for the downslope of the curve is assigned to no one, so it shows up as a surprise in production when it was already predictable on a whiteboard.

I have watched this curve since 1990

I have been building software since 1990. If thirty-six years taught me one thing, it is that every wave of technology sells at its peak of confidence first.

ERP in the nineties was going to put the whole company in order the moment you installed it. CRM was going to make sure no customer slipped through the cracks. Digital transformation was going to reinvent the business. In every case the peak felt just as real, and in every case the downslope came from the same direction: people did not use the system the way the plan assumed, the real processes looked nothing like the org chart, and the last mile of adoption ate the promised return.

What changed this time is the tool, not the curve. The models are better, building is faster, the demos are more convincing. But the downslope is still the same one as always: the organization was not ready for what got built. Anyone telling you the tech is now so good it skips the downslope is standing right on the peak.

The people past the valley are right

The most interesting part of Mannion’s post happened in the comments. J. Patrick McDonald, who advises on AI governance, did not argue with the curve. He argued with where we are standing: he and many of the CTOs he works with hit the bottom of the valley over a year ago and are already climbing out the other side.

And he dropped the line that matters: 90% of those pilots never had a chance to show ROI, because ROI was never considered when the pilot started. The value case was thin to non-existent, built on assumptions nobody validated.

That does not contradict the downslope thesis. It confirms it. If the return was never defined, of course the pilot moved no P&L: nobody knew what it was supposed to move. Jason Duncan put a finer point on it in the same thread: a lot of projects begin with a solution and then go searching for a problem, and by the time someone asks how success will be measured, the project is already underway. Technology accelerates progress, yes, and it accelerates ambiguity too.

That is precisely the decision a discovery makes before a single line of code gets written: what problem, what return, what happens if the honest answer is that it is not worth it. That is not bureaucracy. It is the insurance against the downslope.

What we do about it at IQ Source

When a company asks us for help with AI, the first question is not what to build. It is whether to build, and who will own the downslope when it arrives.

AI Maestro is that discovery. Two months mapping the real operation, not the org chart, to produce three things: a map of how the process actually works, an AI Opportunity Score that says where the return is real and where it is just a pretty demo, and a Go/No-Go gate at the end that decides, with data, whether to build at all. More than once that gate ends up recommending not building yet, or building far less than the client came in asking for.

What that discovery really does is give the downslope an owner before it exists. The resistance, the edge cases, the last mile: all of it gets surfaced during discovery, on paper, while changing course is still cheap. It is the same discipline I wrote about when I argued that adopting AI is not the same as transforming, and the one I apply when I help a company decide whether AI is even the answer for a given process. Often it is not, and knowing that early is worth more than any pilot.

Peak confidence feels good. That is what makes it dangerous. The question worth asking in your next AI meeting is not “what do we build?” It is the more uncomfortable one: when this reaches production and the resistance nobody mentioned shows up, who, by name, owns that downslope? If nobody raises a hand, you are not ready to build yet. You are standing on the peak.

Give your AI project’s downslope an owner