A founder ships a 60,000-line pull request into production, "gleaming with joy at how much more productive they've become with AI."
The CTO sees the tech debt immediately. And can't say anything.
Because raising it means looking like a "Debbie Downer" while the founder rides an AI high. So the CTO does the only thing left: he signs out of the job. No next gig lined up.
That's one of almost 20 engineering leaders Gergely Orosz talked to who are on their way out or seriously considering it. Six of ten told him they're leaving.
Now hold that against a second fact from the same fortnight. Asana needed to migrate off the testing framework Enzyme — a massive rewrite of test cases, the kind of deep, unglamorous work that "would surely have been kicked down the road." With AI, they finished it in two weeks. Airbnb and Uber reportedly share similar stories.
So here's the paradox, laid bare. AI is finally doing the exact deep technical work engineering orgs always deferred — and the people who'd normally shepherd that work are walking out the door, right as it starts paying off.
It reads like a tragedy. In one reading, the steersmen abandoning ship the moment the engine finally turns over.
It isn't a tragedy. It's the point.
The engineering-leader exodus, in this interpretation, isn't a talent crisis. It's arguably the org chart correcting itself. And to see why, you have to look at what a Director+ role was actually built to sell.
Start with the money. Base salary at Director, VP of Engineering, and CTO levels is often similar to a staff engineer's. The premium was never in the paycheck — it was in equity, and in strategic ownership. You took the seat because you got a larger stake in the company's future and real control over where the engineering org went.
That control had three concrete components. Coordinating large teams. Gatekeeping quality. Owning multi-quarter roadmaps.
Arguably, AI is unbundling all three at once.
Coordination first. One of Gergely's ten reasons for the exodus is blunt: "smaller teams mean less need for leaders." When AI collapses a testing-framework migration from a deferred multi-month slog into two weeks, you need fewer people to run it — and fewer layers of leadership to coordinate those people. The coordination value evaporates with the headcount it was coordinating.
Quality gatekeeping next. A VP of engineering on his way out put it plainly: "Shipping software became all about speed. Finding differentiation with your product in the market is brutal, and speed / go-to-market becomes the biggest differentiator. Craft, quality, and care going into the product are taking a backseat."
The quality gate was the leader's lever. When the org decides speed is the only differentiator that matters, the gate gets propped open — and the gatekeeper is holding a badge that no longer opens anything.
Then the roadmap. This is where the 60,000-line PR does its real damage. It isn't just tech debt. It's a signal that the founder can overrule whatever was agreed about what to build next — on vibes. Feelings the founder has are reason enough.
You cannot own a multi-quarter roadmap when a single founder afternoon with an AI coding tool can rewrite it. The seat that owned direction still exists on the org chart. It just doesn't own direction anymore.
So the leaders best positioned to steer the AI transformation are leaving — arguably because the transformation dissolves the very seat they'd steer from. That reading frames it not as irony but as the mechanism.
Now flip the whole thing over, and you get the same signal from below.
Elena Verna ran a study of 320 people on IC work. The headline: 71% of people — managers included — want at least half their time spent as individual contributors. Strip away compensation, title, and prestige, and the largest single group, 27%, want to do IC work full time.
Among current managers, only 25 of 128 would choose to do 100% management. Of the 35 people in the whole study who wanted management full-time, 31 were already managers or player-coaches. Almost nobody chooses into it from the outside.
Here's the part that matters most. When people named the biggest downside of IC work, 66% said lack of influence.
Sit with that. Influence is the one real thing management buys you that IC work doesn't. It's the reason people tolerated the meetings, the politics, the responsibility for other people's careers.
And influence is precisely what founder-mode strips out of the leadership role anyway.
That's the collision. The one benefit that made management worth the pain is the exact thing evaporating when a founder ships a 60,000-line PR and overrules the roadmap on feelings. The IC gives up influence and gets hands-on work back. The founder-mode CTO gives up hands-on work and gets no influence.
The IC gives up influence and gets hands-on work back. The founder-mode CTO gives up hands-on work and gets no influence.
Why would anyone stay in the second seat? They wouldn't. Six of ten aren't.
There's a market signal underneath this too, and it's brutal. Another of Gergely's reasons for the exodus: AI startups now pay ICs more than non-AI startups pay executives. Arguably, the comp ladder has inverted. The seat with the least influence is no longer even the seat with the most money.
So where does strategic direction go? Because it doesn't vanish. Someone still decides what to build.
It relocates in one of two directions.
Direction one: into the founder. The founder now ships code, sets the roadmap by vibe, and absorbs the strategic ownership that used to live in the CTO seat. That's founder mode taken to its logical end — the founder as the org's only genuine director. It's fast, and it comes with the accountability hole nobody's owning: who's oncall when founder-shipped code breaks.
Direction two: into a new shape entirely. Verna's data surfaces a third group she almost didn't think to study — "player-coaches" who work as a manager and an IC at once. Some love it, some hate it. But that hybrid is the emerging container for strategic direction in a smaller org: the high-impact IC who still shapes what gets built, without a large team to coordinate or a quality gate to guard.
One might call it the HI-C player-coach, though the evidence treats high-impact ICs and player-coaches as distinct groups. It arguably keeps the influence and drops the coordination overhead that AI just made cheap. It's the Director+ value proposition rebuilt for a world where the team is small and the roadmap is fluid.
Both destinations are real. What's not yet decided is which one companies choose deliberately.
Because right now, most aren't choosing. They're letting the transition happen by attrition — waiting until the CTO signs out with nothing lined up, then discovering the strategic-direction function left with him.
The evidence pack has one counterexample worth holding onto. CTOs named Ramp, Stripe, and Notion as places that integrate AI into engineering culture with a growth mindset without forsaking quality. Matt Boyle took the VP of Engineering role at Gitpod specifically after interviewing the employer on whether they genuinely leaned into AI's changes — and had a rewarding experience.
The difference between those companies and the ones bleeding leaders, arguably, isn't AI adoption. Everyone's adopting AI. The difference may be whether they redesigned the leadership seat to match, or left the old seat in place and watched it dissolve under whoever was sitting in it.
The transformation is going to relocate strategic direction either way. Into a founder shipping 60,000-line PRs, or into a deliberately designed player-coach role that keeps the influence AI didn't touch.
The only open question for product orgs is whether you draw that new box on the chart yourself — or wait for the person in the old box to draw the door on his way out.