The Burnout Cascade

I left Meta in June 2021. I'd been working 16-hour days, watching the company scale from 17,000 to over 63,000 employees. The pace was brutal but manageable because the release cycles had a rhythm. Quarterly planning. Semi-annual launches. Yearly reorgs. You could see the wave coming and brace for it.

I left because I was burned out, and because a startup had copied my side project's landing page word-for-word and raised a Sequoia round while I sat inside a company that wouldn't let me compete. But the burnout was the kind I understood. It had edges. You could step away from it.

The burnout happening now is different. It has no edges.

The cascade

AI labs are in a civilization-scale arms race. OpenAI, Anthropic, Google, xAI, Meta AI — all sprinting on compressed timelines because the cost of being second might be existential. Inside those labs, the pressure is immense and well-documented. Engineers shipping frontier models every few months. Research teams publishing at a pace that makes peer review irrelevant. The human cost is visible if you know where to look: attrition rates, quiet departures, the occasional candid tweet from someone who just left.

That internal pressure creates an external release cadence. A new model drops. APIs change. Benchmarks reset. Capabilities that didn't exist 90 days ago are now table stakes.

And then the cascade begins.

Developers absorb the first wave. Your stack shifted. The framework you learned last quarter is already behind. The API you built against just deprecated the endpoint you depend on. You're not falling behind because you're lazy — you're falling behind because the ground won't stop moving.

Educators absorb the second wave. The curriculum you designed in September is outdated by January. Students are using tools that didn't exist when you wrote the syllabus. You're simultaneously expected to integrate AI into your teaching and to police its misuse, and nobody trained you for either.

Designers, writers, product managers, support teams — they absorb the third wave. The workflows they mastered keep dissolving. Skills that took years to build get partially automated every quarter. Not fully automated, not yet, but enough to create constant anxiety about relevance.

The source of the stress wave is the labs. Everyone else is downstream of their sprint cadence.

When burnout had a season

Before AI compressed the cycle, burnout in tech followed seasons. You'd push hard for a launch, recover during a slower quarter, then push again. Social media release cycles were measured in months. Product cycles in quarters. Platform migrations in years.

I cycled through burnout at Meta like people cycle through pants. It was almost the default operating mode. But the cycles had shape. You could predict the peak, plan for the trough, build recovery into the rhythm. The company even had "recharge" — a full month off every five years of service, on top of normal PTO and company holidays. Institutionalized acknowledgment that the pace was unsustainable, with a structured way to recover from it.

The AI cycle has no troughs. There is no recharge week when your competitor just shipped a reasoning model that obsoletes your roadmap. There is no "slower quarter" when capabilities compound monthly. The release cadence at the labs sets the clock for the entire industry, and that clock has no off-position.

The flip

When I left Meta, my thesis was simple: the future belongs to smaller teams that punch far above their weight. I'd watched a 63,000-person company move slower than two engineers in a garage. I'd watched my own side project grow entirely on its own during the pandemic while I was stuck in meetings about meetings. The math seemed obvious — smaller, faster, more capable per person.

I didn't think it would happen this fast.

The curve went vertical. A solo founder running AI agents can now produce the operational output of a small team. I do it. $600 a month in API costs, seven agents, daily standups, weekly reports, content drafts, deployment monitoring. The work that used to require hiring three to five people now runs on cron jobs and context windows.

But the flip has a dark side. If one person can do the work of five, what happens to the other four?

Their value model gets repriced.

Execution — the thing most knowledge workers sell — is getting cheaper by the quarter. Writing, analysis, triage, coordination, monitoring, reporting. Each one a little more automatable with each model release. Not gone. Cheaper. Which, in labor economics, is almost the same thing.

What compounds differently

The skills that aren't getting cheaper: judgment, taste, relationships, distribution, cultural fluency, the ability to know which problem is worth solving. These compound with experience. They get more valuable as execution costs collapse, because the bottleneck shifts from "can we build this?" to "should we build this?"

I spent five years at Meta watching brilliant engineers build things nobody needed because the organizational machine incentivized shipping over thinking. The bottleneck was never execution. It was always judgment. We just couldn't see it because execution was expensive enough to feel like the hard part.

Now execution is cheap and getting cheaper. The illusion dissolves. What remains is the question that was always underneath: do you know what matters?

I'm still in the cascade

I'm not writing this from a position of comfort. I cycle through burnout like it's weather. I left big tech, burned out on the startup grind, burned out on the infrastructure migration, nearly gave my product away twice, and I'm still here — shipping at midnight, debugging auth at 2am, running seven agents that occasionally restart my production services without asking.

The difference between this burnout and the Meta burnout is that this one has purpose. I chose this pace. I set the clock. The agents work for me, not the other way around.

That's the invitation. Not "quit your job" — most people can't, and shouldn't, based on a blog post. But: look at where your value actually comes from. If it comes from execution speed, from the sheer ability to produce output, the pricing on that is changing fast. If it comes from knowing which output matters, from relationships that can't be API'd, from taste that took a decade to develop — that's the stuff that compounds while everything else deflates.

The labs will keep sprinting. The release cadence won't slow down. The cascade will keep flowing downstream.

The question is whether you're standing in the current or redirecting it.