AI Psychosis
The Conversation CTOs Aren’t Being Coached to Have.
A founder watches an AI tool do in minutes what used to take a sprint, and assumes the rest of the org can move the same way. The CTO knows better, and has to say so.
The instinct is understandable: Why aren’t we doing this already?
Maybe the expectation is that engineering should now be twice as fast, or the company should need fewer engineers, or there’s simply another company being held up as proof that it can be done.
The CTO inherits the gap between the demo and reality.
The term “AI psychosis” isn’t mine. Box CEO Aaron Levie used it in a widely shared post earlier this year, and it stuck because it named something real: leaders far enough from the work that a good demo looks like a solved problem.
That’s where the interesting problem starts. The AI works. The productivity gains are real. But underneath the speed, code is accumulating that nobody fully understands, architectures are diverging, and teams outside engineering are shipping things that eventually land on engineering’s desk.
This is what I’m calling AI psychosis tech debt: the gap between what leadership believes AI should make possible and what the organization can actually sustain in production.
And the hard part for the CTO isn’t identifying the debt. It’s having the conversation with the person who is most excited about creating it.
To be clear, none of this is an argument against AI. Almost every company we advise has adopted AI-assisted development in some form, and a lot of our own work is helping them mature that Agentic SDLC, not slow it down. We’re all in on this too. Agentic software development is the default today, and the gains are real, and in most cases measurable.
What’s getting far less attention is what’s building underneath those gains.
What We’re Seeing
Over the past nine months, the majority of work coming through Hoola Hoop has touched some version of this problem.
It usually comes in one of two ways.
The first is the CEO or investor. The organization feels too slow. AI should mean the team can move much faster. There’s usually another company being held up as proof that it’s possible, although very little is known about what’s actually happening under that company’s hood.
The second is the CTO. They’re worried about what that speed is producing. More and more software is being built outside engineering. Quick wins have turned into inconsistent architecture, security issues, and code that engineering now has to own.
We’ve spent more than one CTO Roundtable talking about exactly this. It’s not an edge case anymore. It’s becoming a normal part of the CTO job.
Why AI Psychosis Becomes Tech Debt
The mechanism is pretty simple.
Someone sees an impressive demo. They form a belief about what the organization should now be able to do. That belief quickly becomes an expectation, and the expectation becomes a cost or productivity target before anyone has tested it against production reality.
The gap is increasingly measurable. Research from Veracode and the Cloud Security Alliance points to significant security and architectural problems in AI-generated and AI-assisted code. The exact percentages matter less to me than the pattern: the code can look finished long before it is actually ready for production.
That’s the bill that comes due after the room that applauded the demo has moved on.
And the debt doesn’t come from one place.
Where the Debt Comes From
Some of it is built outside engineering entirely. Marketing, operations, product, and others can now use AI tools to build working software without engineering ever seeing it.
The rest is built inside engineering, where teams are moving faster without the context, guardrails, standards, and architectural discipline that speed actually requires.
Neither is an argument against the tools.
It’s what happens when the speed of building changes faster than the operating model around it.
Four Patterns Behind AI Psychosis Tech Debt
Across the CTOs we coach and advise, similar patterns keep showing up. Recognizing which one you’re dealing with is often the fastest way to get unstuck.
The Common Thread
All four patterns come back to the same gap: confidence formed at a distance from the work is being treated like confidence formed by doing the work.
A demo can create enormous confidence very quickly. So can an AI assistant that tells you your idea is sound. Neither tells you what happens when the software hits real users, real data, real security requirements, and the rest of the system.
That’s not a personality flaw in the founder or CEO. It’s a structural problem, and it’s what turns into AI psychosis tech debt if nobody names it early. It is one the CTO has to learn to manage.
How to Handle the Conversation
The CTOs who navigate this well aren’t relying on better prompts. They’re building the operating model that was supposed to exist before the tools got this capable.
The Conversation Itself
This is where the CTO’s job gets harder.
You can be completely right about the technical problem and still handle the conversation badly.
We’ve talked about this repeatedly in our CTO Events & Roundtables The goal isn’t to shut down the people building outside engineering. They’re usually solving a real problem, and they’re solving it quickly. Trying to stop them often just pushes the work somewhere with even less visibility.
What works better is bringing them inside the operating model.
Define what can be built without engineering review, what needs a checkpoint before it touches customer data, and what production-ready actually means. Give people approved patterns and tools before three different versions of the same thing appear across the company.
That’s a partnership, not a policing effort.
And when you need to push back on the CEO or board, make the conversation about the tradeoff, not about who’s right.
The tech debt is usually manageable. What’s harder to manage is being the person who has to say so, to the person who’s most excited about the thing creating it.
Questions to Ask Yourself
If this conversation is coming for you this quarter, these are worth working through before you walk into the room.
- Do you have a number that demonstrates the problem, or are you relying on a general sense that something is off?
- When you last raised the concern, did you lead with the problem or with a plan? Which one did the room respond to?
- Have you rehearsed this conversation with someone before you needed to have it for real?
A Final Thought
AI psychosis tech debt isn’t going away. It will show up again next quarter under a different name, attached to a different demo.
The question is whether the CTO has practiced this conversation before it happens.
Because the hard part isn’t knowing that technical debt exists. Most CTOs already know that.
The hard part is telling the CEO or board, calmly and with evidence, “Yes, we can move faster. Here’s what it’s costing us, and here’s what I’d like to do about it,”
That’s a skill. And like most leadership skills, it gets better with practice.
If AI psychosis tech debt is the specific flavor you’re dealing with, you’ll find more of our thinking on it in the CTO Coaching articles on our site.
Ready to talk about CTO coaching with Leigh?
Book a 30-minute introductory call to explore whether coaching is right for you.
