Both camps are right about a real, existential threat. They've just stopped being able to see each other's threat — and that blind spot, not the technology, is what's breaking teams.
From the essay by Charity Majors · charity.wtf →
The industry has split into two tribes over AI-assisted development, and they've largely stopped talking across the divide. Majors' move is to refuse the usual framing where one side is wise and the other is foolish. Both sides are reacting to a genuine, existential threat. They're just running from different ones.
This isn't smart-vs-dumb. It's two true things on two different clocks: the enthusiast's danger arrives quickly and from outside; the skeptic's arrives slowly and from inside. A team that only hears one clock will mismanage the other.
Here's why the two camps can't just "look at the evidence" and agree: they're looking at different parts of the same shipped work. Flip the lens on one AI-built feature and watch the story invert.
Majors doesn't literally toggle a UI; this widget dramatizes her point that wins and costs are invisible to each other. The enthusiast genuinely never sees the cleanup; the skeptic genuinely never sees the win.
If both sides can see the truth, why doesn't reality converge? Because the feedback is asymmetricalWins and costs travel through different channels at different volumes — so the public record systematically over-represents one and hides the other.. Wins get broadcast — conference talks, demos, celebratory threads. Costs surface quietly, later, in private: maintenance meetings, on-call gripes, a Slack DM.
Ship a few AI features below. Every ship that works gets posted to the public feed instantly. The downstream cost, when there is one, lands in the maintenance channel — which, by default, you can't see.
Ship 6–8 features with the maintenance channel hidden, read the narrative, then flip it on. The public record said "AI is pure magic." The hidden ledger was accruing the whole time. Neither camp is lying — they're each reading a different, incomplete feed.
So who's actually right about outcomes? Majors' answer, backed by the data: it depends on what you already had. AI doesn't decide whether you win or rot — it multiplies whichever you were already doing.
"AI is an amplifier… it magnifies the strengths of high-performing organizations." — 2025 DORA report, cited in the essay
Drag the engineering discipline knob — testing, observability, fast feedback loopsThe connections between an action and its consequence that let a team learn. Tight loops catch a bad change in minutes; loose loops let it rot for months.. The raw AI speed-up stays the same for everyone. Watch what that same speed-up turns into.
Darragh Curran's team at Fin (formerly Intercom) 3×'d output in nine months while defects and downtime went down. Not because of the AI alone — because they had "exceptionally high engineering discipline, fast feedback loops, and a culture of experimentation." Crank the knob to the top: that's the regime where the amplifier pays off.
Now drag it below the midpoint. Apparent output stays high — that's the conference-talk number — but net progress goes negative. That's the skeptic's entropy: shipping faster into a system that can't absorb it.
Majors' prescription isn't "pick a side." It's to rebuild the shared reality the asymmetry destroyed. The core reframe: stop treating the disagreement as a debate to win and start treating it as a system to diagnose.
Each side defends its position. The goal is to be right. Costs and wins stay weaponized — evidence to score points, not to learn from. Nobody updates.
Treat the gap as a system to fix. Ask what conditions would make the other side's fear stop coming true — then go build those conditions. Now the disagreement is actionable.
Flip the switch above. The rest of her playbook all serves that one shift:
→ Tell complete stories — report the cost and the win, in the same shared forum, so the record stops being asymmetrical.
→ Build credibility through expertise — earn the right to an opinion by actually understanding the new tools.
→ Keep feedback loops intact — preserve the connection between a change and its consequence; that's what lets the team learn instead of guess.
→ Lead peer-to-peer — trust is built sideways between engineers, not handed down as a mandate.
Leaders are steering between two real hazards at once: missing the AI opportunity and grinding the team into exhaustion. You can't mandate your way through that. You acknowledge both threats are legitimate and apply engineering discipline to the transition itself.
An engineer watches a dazzling AI demo at a conference and concludes their skeptical teammates are just being Luddites. By Majors' argument, what is the engineer most likely missing?
Enthusiasts and skeptics are both right, racing different clocks — but asymmetrical feedback hides each side's reality from the other. AI only amplifies the engineering discipline you already had. The fix isn't winning the debate; it's restoring a shared, complete record and asking "what would it take?"