← Back to Posts

AI Vampires and the 100x Org: When Silicon Valley Lost the Plot

#ai , #opinion , #industry , #andreessen , #clickup , #burnout , #analysis

The Vampires Are Out

Marc Andreessen went on a podcast and said the quiet part loud. AI didnt replace programmers. It turned them into vampires.

Heres the actual quote, from the a16z podcast with Erik Torenberg (also recycled across his recent Joe Rogan appearance):

“The opportunity cost of going to sleep is too high, because if you go to sleep, you wont be with your 20 AI coding agents.”

He describes them as “completely exhausted, but euphoric. Theyre thrilled.” Huge bags under the eyes. No sleep. Twenty agents going at once. According to him, leading-edge programmers are now 10-20x more productive than a year ago, and instead of using that productivity to, you know, live a normal human life, theyre cranking harder than ever.

The man who wrote “Software is eating the world” fifteen years ago is now telling us that software is, in fact, eating the developers writing it. Cool. Great. Very on brand.

Who Is Marc Andreessen and Why Does This Matter

Quick recap for anyone who has been touching grass. Andreessen co-founded Netscape in the 90s, then Andreessen Horowitz (a16z), one of the most influential VC firms in Silicon Valley. They were early in Facebook, Airbnb, Coinbase, GitHub. He has been the loudest voice on the “techno-optimist” side of the tech debate for a decade, and his portfolio has billions of dollars riding on AI tooling being huge.

So when he frames sleep-deprived coders as the new normal, hes not a neutral observer. Hes describing the working conditions inside companies hes invested in. Take that as you will.

The “AI vampire” term itself is real and spreading. Search it and youll find Polymarket, Fortune, podcast notes, X posts, all using it within the last few weeks. Like it or not, it has caught on.

The Dopamine Loop Nobody Talks About

Heres the thing that the productivity numbers always skip. When an agent ships code while you watch, your brain treats it like progress. Even when nothing useful is shipped, you feel like you built something. Thats not engineering, thats slot machine logic.

I notice this in my own workflow. I open three Claude sessions in three tabs, half the work isnt code, its prompt babysitting. And every time one of them finishes a task I get the little hit. Even when the output is half wrong and Ill have to rewrite it. The hit comes first, the verification comes later. If the verification ever happens.

Multiply that by 20 agents and you can see how someone ends up at 4am wondering where the night went, convinced they shipped a startup. They didnt. They shipped four half-finished side projects nobody asked for. But the dopamine was real.

The actual hard problem in software, the one nobody likes to talk about, isnt writing code faster. Its distribution. Its product-market fit. Its convincing humans to switch from the thing they already use to the thing youve built. None of that gets easier because your agent fleet runs at 3am. You can build a better Airbnb clone in a weekend now. So what. Airbnb has the listings, the brand, the SEO, the trust. Your weekend clone has nothing.

But “build faster” is what AI sells well, so thats the story everyone tells.

Then ClickUp Said Hold My Beer

While Andreessen was talking about vampires on a podcast, ClickUp turned the metaphor into an HR policy.

On May 21, 2026, CEO Zeb Evans posted a memo on his personal X account announcing a 22% workforce reduction. Roughly 290 people gone out of about 1,300. The framing was the unusual part. This wasnt a cost-cutting layoff, Evans insisted. This was a structural redesign around something he called the “100x org.”

The actual claims in the memo, paraphrased from the public coverage since the full text isnt easily linkable:

  • ClickUp is restructuring so AI agents outnumber human employees roughly 3 to 1 (so, around 3,000 internal agents).
  • The remaining workforce gets compressed into three categories: “builders” (engineers and PMs directing agent fleets), “system managers” (people who automate other roles and then own the resulting systems), and “front-liners” (customer-facing staff).
  • The savings from cutting 22% of headcount dont go back to the income statement. They get redirected into new salary bands that top out at $1 million per year, available to anyone who can demonstrate a “100x impact” through AI systems.
  • The most quoted line: “The best engineers are not writing code anymore. They are directing agents that write code.”

So yeah. ClickUp is running a corporate Hunger Games. Survive the cut, prove you can move the needle 100x, get a million dollars. Allegedly.

Lets Talk About That 100x Number

100x is doing a lot of work in this memo. Not 2x. Not 10x. One hundred times more impact than the baseline, whatever the baseline is. Evans uses the number repeatedly, which means hes serious about it, which is where I start having problems.

A 100x increase in code shipped is plausible if you redefine “shipped” loosely enough. A 100x increase in features users actually want, that get adopted, that dont break the existing product? Show me the team that has done this. Ill wait.

ClickUp is a real product with real paying customers. Real users dont want their tool to change 100x. They want it to keep working. They want the button to stay where the button was. The product surface of a SaaS tool has a natural rate of acceptable change, and its way lower than “100x.” Try to push past it and your users dont feel innovation, they feel motion sickness. Then they churn.

This is also where Im naturally skeptical of “we cut 22% to invest more in the rest” memos. Sometimes thats true. Sometimes its a layoff dressed up in optimism because optimism reads better in press coverage. The line between “strategic restructuring” and “we ran out of runway and the AI story sells” is thin and well-traveled.

I dont know which one this is. Im not inside ClickUp. But the $1M salary band thing has a real “the beatings will continue until morale improves” energy to it. If you have to incentivize people with million-dollar salaries to extract 100x impact, you might be paying them to do something thats not actually possible.

What the Job Market Actually Says

Now lets bring in the data, because both sides of this story love to cherry-pick.

The bearish case. Goldman Sachs economist Elsie Peng published a U.S. Daily Note on April 6, 2026, based on actual payroll records. AI eliminated about 25,000 jobs per month through substitution while creating about 9,000 per month through augmentation. Net loss: 16,000 jobs per month, sustained for about a year. Entry-level software developer employment for the 22-25 age bracket is down nearly 20% since 2024. The new grads are getting hit hardest.

The bullish case. The FRED Indeed software development job postings index sat at 73.10 as of May 15, 2026. Lower than the 2022 ZIRP peak by a wide margin, but trending up in the last twelve months. So job postings exist, theyre recovering, theyre just not anywhere near the highs that older devs remember.

The thing nobody wants to say. A lot of the “AI replaced X jobs” narrative is upstream of a different story. As reported by Fortune and Tom’s Hardware, Microsoft started canceling many of its direct Claude Code licenses because employees burned through tokens way faster than expected. Uber blew through its entire 2026 AI coding tools budget in four months. Agentic AI can use up to 1,000x more tokens than a normal LLM query. Whats actually happening at many companies isnt “AI is now cheaper than humans.” Its “weve laid people off to free up budget for AI compute that turned out to cost more than the humans did, and now were stuck.”

That last one should be on the front page of every tech outlet for a month. Its not, because it doesnt fit either narrative neatly.

The Two Stories Are the Same Story

Andreessen and Evans are saying the same thing from two angles.

Andreessen is describing the worker side: programmers cant sleep because they feel they have to maximize their agent fleet, or else they fall behind.

Evans is describing the company side: were going to formalize that pressure into our org chart and our comp bands. Output 100x or be one of the 22%.

The middle of this graph is one giant treadmill. If everyone is supposed to be 100x, then 100x is the new 1x. Then someone redefines the goalposts and we all start over. Thats not a productivity revolution. Thats a wage and burnout spiral with extra steps.

And the thing is, neither side has the receipts to prove that this 100x output actually exists in any sustained, measurable way. Theres a lot of “the best engineers are now ten thousand times more productive” rhetoric and very little “heres our actual revenue per engineer YoY and heres how it correlates to AI tool adoption” data. When the numbers do show up, they tend to look like Microsofts: the tools cost more than expected, the gains are real but bounded, the burnout is also real.

My Take, For Whatever Its Worth

I dont want to dunk on either of these guys too hard, because theres a real signal underneath all the noise.

Andreessen is right that some programmers are clocking absurd hours with AI tools and feeling productive doing it. Ive done it. You probably have too. The tools are genuinely impressive and the temptation to keep them spinning is real. The “vampire” framing is annoying but its describing a real pattern.

Evans is right that AI changes what roles look like. The bottleneck in a lot of teams really has shifted from writing code to orchestrating systems, reviewing output, and figuring out what to build. The people who are good at that part are getting more leverage. Thats true.

Where I get nervous is when both of those reasonable observations get inflated into “we should all be 100x” and “sleep is a waste.” Because thats not how humans work, and its not how products work either. Burnt-out engineers ship worse code, not more code. Users dont want a product that mutates 100x. And companies that turn their org charts into Hunger Games tend to lose their best people first, not last, because the best people have options.

Maybe Im wrong. Maybe Evans pulls it off, ClickUp 6.0 is incredible, the survivors take home a million dollars each, and we all look like idiots in two years. Stranger things have happened in this industry. Im not going to pretend I can predict it.

What I do think is worth saying out loud is this: the “AI productivity miracle” narrative is currently outrunning the data supporting it. Were making strategic decisions, including who gets to keep a job, based on stories told on podcasts and X memos rather than measured outcomes. That gap will close eventually, one way or the other. I would rather we close it before we burn through the careers of an entire generation of developers chasing a 100x dragon that may or may not exist.

Build less, sleep more, ship things that matter. Boring advice. Probably right.

References