The Invisible Burnout of the AI-Augmented Developer

The Productivity Paradox

Since we started building Catena with AI agents, our velocity has exploded. We ship features in 3 days instead of 15. We test product hypotheses in a few hours. We poc, deploy, iterate… two of us, with the firepower of a team of seven.

But there's a downside I wasn't talking about. A downside that took me months to identify, because it looks like nothing we know. No crunch. No all-nighters. No visible overwork. Just a brain that never stops running.

I'm going to talk about the invisible burnout of AI-augmented development (for developers, PMs, POs, etc.). And why, paradoxically, it's exactly the opposite of what we've been preaching for years with Agility.

The Agile Anti-Pattern Nobody Sees

We repeat it in every training, every workshop, every team ritual. The fundamental principle of agile productivity is flow: do smaller actions, but end-to-end. Finish one subject before starting another. Limit work in progress. Kanban's famous WIP limit.

Why? Because context switching has an enormous cognitive cost. Every time we change subjects, the brain needs an adaptation phase to get back into context. Cognitive psychology studies talk about 15 to 25 minutes to regain deep concentration after an interruption. If we jump from subject to subject all day, we never reach that state. We stay on the surface. We exhaust ourselves.

This is exactly the foundation of Lean: reduce waste linked to work-in-progress, mental queues, multitasking. Toyota understood it for production lines. Agility transposed it to software development.

The AI-Augmented Dev Workflow

When you develop with an AI agent, the workflow changes radically. And that's where the trap snaps shut, almost without noticing.

You prompt a feature. AI works. While it generates, you don't sit idle. You move to another subject. You draft a spec. You review a design. You launch another prompt on another context. AI comes back with a result. You test, adjust, re-prompt. You return to the first subject. You've lost the thread. You re-read. You re-contextualize. You re-prompt. Meanwhile, a third idea has germinated. And this feature too, hey, it would be easy to do. And this one as well.

The brain no longer processes one subject at a time. It constantly juggles between 5, 8, 10 different contexts. Not by choice, but because the tool allows it. Because AI's speed creates an irresistible pull.

The Invisible Mental Load

What makes this phenomenon particularly dangerous is that it's invisible. The classic burnout signals aren't there. No overtime. Going home at a reasonable hour.

But the brain hasn't clocked out. It keeps running on the 10 open subjects. It loops back on the unresolved bug from prompt 3, the architecture from prompt 7, the odd result from prompt 5, the new feature (the new features) that germinated during the day. At night, it compiles. On weekends, it iterates in the background.

This is a form of overload that can't be measured in hours. It's measured in simultaneously active contexts. And that number, with AI-assisted development, has multiplied by 3 or 4 without anyone noticing.

The symptoms are diffuse: difficulty concentrating on a single subject, feeling of never "finishing" anything, mental fatigue disproportionate to time worked, irritability, impression of running without moving forward. Nothing spectacular. Nothing that triggers an alert. But a wear that accumulates.

What We Changed (and What I Recommend)

I don't claim to have the definitive solution. But here's what we put in place at Catena, drawing precisely on the agile principles we had forgotten (ironic for me 🫣).

• Limit WIP, even with AI. We impose a maximum of 2 parallel subjects per person. Even if AI could handle 10. We are the bottleneck, not AI. Better to accept that than pretend otherwise.
• Timebox prompting sessions. We use our own tools (Catena's sequences) to structure our development sessions. 45 minutes on one subject, break, decision: continue or move on. But no juggling.
• Separate generation time from validation time. Instead of launching 5 prompts in parallel and validating on the fly, we batch: one generation phase, then one review phase. The brain stays in one mode at a time.
• Protect time without AI. Moments where we think without prompting, without waiting for a result, without the reflex of "hey, I could ask AI about this". Counter-intuitive when you have such a powerful tool at hand. But necessary.
• Bonus: An instant ideas backlog. Ideas that come during prompting, review, or testing phases aren't acted on immediately. They go to the parking lot (Lean/agile concept 😊) then sorted at end of day.

A Collective Challenge, Not Just Individual

This topic goes beyond Catena. Looking at the tech ecosystem, we see entire teams switching to AI-assisted development without any reflection on the cognitive impact.

In the months and years ahead, we're going to see a new form of burnout emerge. Not the classic burnout linked to work overload. A burnout linked to context overload. Developers, product managers, founders who work "normal" hours but whose brains are constantly overheating because they're managing 10× more flows than before.

Companies that take this seriously will have an advantage. Not just for their teams' well-being, but for the quality of what they produce. Because an overheating brain doesn't make good decisions. It doesn't do good design. It doesn't build good products.

AI Is Still Formidable… But We're Still Human

To be clear: I have no regrets. AI-assisted development has transformed how we build Catena. We're doing things that were impossible two years ago. The productivity is real. The gains are concrete.

But productivity without sustainability isn't productivity. It's debt. Cognitive debt. And like all debt, it eventually has to be paid.

This article was written by Aymeric Proux, founder of Catena. To discuss facilitation or our tools, find me on LinkedIn.