We talked to a founder last month who had shipped eleven features in two weeks using AI coding tools. Eleven. She looked exhausted. When we asked which features her customers had requested, she went quiet for a long time. "Maybe two," she said. The other nine were things she could build, so she did. Her product was more capable than ever. Her churn rate was climbing.
That conversation crystallized something we have been watching across the network for months. The AI power users — the people going deepest, shipping fastest, producing most — are burning out first. Not despite their productivity. Because of it.
Why are the heaviest AI adopters burning out?
The assumption was clean: AI tools reduce workload, freeing time for higher-order thinking. The reality, for a significant cohort, has been the opposite.
What AI amplifies is throughput, not clarity. If you know exactly what to build and why, AI is a force multiplier. Genuinely extraordinary. If you do not, AI is an exhaustion multiplier. You produce more, ship more, context-switch more — with no increase in the strategic direction that would make the output mean something.
The burnout is not from working harder. It is from working faster in a direction you have not verified, and accumulating the quiet cognitive debt of sensing — at some level you cannot quite name — that the output does not add up to progress.
We have felt this ourselves. There were weeks early on when our own output tripled and our clarity did not keep pace. The feeling is specific: a kind of productive vertigo. The screen is full of shipped work. The satisfaction is absent.
What do 1.7x more bugs and 2.74x more vulnerabilities tell us?
The production data confirms what the burnout data suggests. Research on AI co-authored code reveals two numbers that should stop anyone mid-sprint:
- 1.7x more major bugs compared to human-only code
- 2.74x more security vulnerabilities compared to human-only code
Nearly three times the security vulnerabilities. The mechanism is straightforward: AI generates code faster than people can review it. Speed outpaces judgment. Output outpaces understanding.
When 63% of vibe coding practitioners have zero programming background, code review is not merely difficult — it is structurally absent. The code works. Usually. But the security implications, the architectural debt, and the maintenance burden are invisible to the person who prompted it into existence. They cannot see what they do not know to look for.
This is the inverted bottleneck expressed as a quality crisis. The question is no longer "can we write the code?" It is "do we understand what we are shipping?" The data says: often, no.
What exactly is dark flow?
Mihaly Csikszentmihalyi described flow as deep engagement, loss of self-consciousness, intrinsic reward. Generally positive. The psychological state of peak performance.
Dark flow is its shadow.
Same surface characteristics — absorption, productivity, vanishing sense of time. But the alignment component is missing. You are deeply engaged in activity that is not connected to a direction that matters. Optimizing metrics that do not measure what counts. Busy in a way that feels like progress but does not compound into it.
Productive without purposeful. That is the short version.
The anticivilization trained people for dark flow without naming it. Follow instructions. Hit targets. Optimize metrics someone else defined. The training produces people who are exceptionally good at sustained effort within given parameters. Remove the parameters — as the inverted bottleneck does — and the same training produces sustained effort toward nothing in particular.
Dark flow is not laziness. It is the opposite: intense, skilled, sustained work aimed at the wrong target. And AI tools remove every friction point between intention and output. If your intention is misdirected, you now arrive at the wrong destination faster than ever before.
What happens in the brain during chronic misdirection?
There is a physiological layer to dark flow that makes it self-reinforcing, and understanding it changed how we think about the problem.
Amy Arnsten's research at Yale on the prefrontal cortex (PFC) under stress is the key. The PFC handles long-term planning, strategic evaluation, impulse control — the functions that provide direction. Under chronic stress, the PFC goes partially offline. The brain shifts resources to systems optimized for reactive, short-term responses. You get better at responding to what is in front of you. Worse at evaluating whether your trajectory is correct.
The cycle:
Zero-sum pressure creates chronic stress. Stress suppresses the PFC. Reduced PFC function impairs direction. Without direction, AI-amplified capability produces more output at higher speed — aimed wrong. Misdirected output generates more stress. More stress further suppresses the PFC.
This is not a willpower problem. It is a neurological feedback loop. Dark flow is what happens when you hand regen tools to people operating under degen conditions. The tools amplify whatever state you bring to them. If that state is directionless stress, they amplify directionless stress.
What does dark flow look like at the systemic level?
Noise. Vast, accelerating noise.
When millions of people produce at AI speed without clear direction, the aggregate effect is a supply explosion that overwhelms every discovery mechanism. More apps that address problems nobody has. More features nobody asked for. More projects launched into a void.
Lovable's 100,000 new projects per day — mentioned last week — look different through this lens. The vast majority will never reach a person who needs them. Not because they are bad. Many are technically sound. They are undirected. Built on assumptions about need that were never tested, with tools that never asked whether the assumptions were correct.
More supply is not more value. Supply measures what exists. Value measures what matters to someone. Without the step that connects production to genuine need, the two decouple entirely. We are watching them decouple in real time.
How does the Genius framework address this?
We did not build the Genius framework — Current, Desired, Actions, Results — as a productivity tool. The distinction matters and we want to be precise about it.
Productivity tools optimize the Actions step. They help you do more, faster. They assume you already know what to do and why. In a world of dark flow, that assumption is the problem.
The Genius framework reverses the sequence:
Current. Where are you actually? Not where you wish you were. Not the version you presented to investors. Assessed honestly, with your eyes open.
Desired. Where do you want to be? Not which metric are you chasing. Not what your competitor is doing. What outcome, if you achieved it, would represent real progress — for you, specifically.
Actions. Given the gap between Current and Desired, what actions close it most effectively? This is where AI and production tools become genuinely valuable — after direction is established.
Results. What actually happened? Not what you planned. Not what you hoped. What measurably changed.
The framework is deliberately simple because the problem it addresses is not a complexity problem. Dark flow does not come from insufficient tools or inadequate information. It comes from skipping the direction step. The anticivilization taught people to start at Actions. Do something. Do it faster. Do more of it. The Genius framework insists you start at Current.
We are still learning how this works at scale. We do not have all the answers about what direction infrastructure looks like for different people in different contexts. What we can see clearly is the shape of the problem: the world is full of amplified capacity and short on verified direction. Every system that helps people establish direction before they start producing is working on the right problem.
Direction is the scarce resource. Not capability. Not tools. Not funding. Not credentials.
Without direction, more tools create more noise. With it, every tool becomes a lever.
The dark flow breaks when you stop and ask: where am I, actually? Where do I want to be? That pause — the one that feels like lost productivity — is the most valuable thing you can do right now.
This is Episode 8 of the Superpuzzle Developments series. Next week: "The domino effect" — the mathematical structure that makes each shift toward cooperation make the next one easier.