Big Teams Are Killing Startups

[00:00:00]Two founders, same investor, same room.

[00:00:04]The first one leans back, proud. We just crossed 400 people and we're hiring 30 more engineers this quarter. The second one almost embarrassed. There's 15 of us and we're not hiring anymore. 10 years ago, I know exactly who's getting the check. The big team. Every single time the little guy gets a pat on the head and a lecture about thinking bigger. But sit down those same two founders across from me today and everything's backwards. The big team, red flag. The tiny one, that's the story every investor is chasing because something snapped in the last 18 months. And almost nobody has said it out loud. The one number that used to scream winner, how many people you employ, now quietly means the exact opposite. So stick with me because once you see how this happened, you will never unsee it. Let's go.

[00:01:06]Let's start with why big teams used to be the right answer because it was not stupid. In fact, many investors believed it to be gospel. In 2015, Reed Hoffman, the founder of LinkedIn, gave a lecture called Blitzcaling, which later became a book. Bill Gates wrote the forward and the thesis in Huffman's own words was to prioritize speed over efficiency. That means to deliberately spend capital inefficiently and hire inefficiently because being the first to scale mattered more than anything [music] you wasted getting there. He even gave you the scoreboard. His test for whether you were on track was whether someone would write you a check for 50, 100, $200 million. Money raised, people hired.

[00:01:51]That was the winning model. In fact, the venture capital darling of the day, Uber, raised about $9 billion and torched it on growing as fast as possible. And the book holds them up as the model. And well, to be honest, it worked for 20 years. The logic was sound because two resources were genuinely scarce. Engineers were expensive and slow to find. Capital was a weapon your competitor might get first. So you hoarded both. An army of employees was the moat of businesses. A 500 engineer org took years and hundreds of millions to replicate and that alone kept challenggers out. Every incentive in the industry lined up behind it. Founders bragged headcount. Journalists used it to size companies. [music] Investors reported it to their own backers as proof the money was working. We grew the team to 400 was momentum in every board deck everywhere. But today all of that is falling apart. Step one, the code writes itself. The first domino is the obvious one. AI now handles a huge share of engineering output. But everyone says that without looking more closely at what that means for investing and company growth. At Cognition, the company behind the AI engineer Devon, they state 89% of the code they ship is written by their own product. Even anthropic claims the same thing. In a blog in May 2026, they said 80% of their code is written by Claude and now effectively closes building itself.

[00:03:29]That's crazy. Shopify CEO put out a memo telling every team they must prove AI cannot do a job before they are allowed to hire a human for it. But raw coding speed is the smaller half of the story.

[00:03:44]The bigger half is a law from the 1970s.

[00:03:47]Brook's law from the software engineer Fred Brooks says that adding people to a late project makes it later. That basically means that every person you add multiplies the communication paths.

[00:04:00]A 15 [music] person team has about a hundred possible conversations, but a 200 person company has nearly 20,000.

[00:04:09]Most of the energy in a big organization is not spent on building anything. It is spent on alignment, meetings, threads, signoffs, the daily work of keeping 200 brains pointed the same way. So what we have today is a whole class of jobs existing purely to carry that load. If you work for a big corporation, I am sure you see this all day. The people who translated an executive's idea into a ticket, who turned the meetings into specs, who routed information between teams. human routers. That is the layer AI literally deletes. When the model holds the code base, the context and the road map all at once, any and all translation work more or less disappears. This is why a small team using AI is faster twice over. The machine writes the codes and there's almost no coordination tax left to pay.

[00:05:07]This means less meetings, less employees, and more building. Now, the company's Oric charts of tomorrow won't look like the companies of today. But more importantly, here is what that looks like in the numbers, and it is not subtle. The old benchmark for a well-run software company was 200 to $300,000 of revenue per employee. The median across private SAS is closer to 130,000. That was the physics for two decades and it is true for many of the now public companies from that era of venture capital. But the new wave is very different. Lovable out of Stockholm reached $400 million in annual recurring revenue with 146 full-time employees.

[00:05:54]That is $2.7 million per person, roughly 20 times the old median. Gamma, the AI presentation designer, serves 50 million users with a team of about 30 and has been profitable for 15 straight months.

[00:06:10][music] More interestingly for Gamma, the founder deliberately caps the team because he believes 50 focused people beat 200. And there are plenty of other examples, too. Bold, for example, when that had its hot moment, went from 0 to 20 million in revenue in 2 months with 15 people. Midjourney has built a business worth around $10 billion with a team that fits in one conference room on zero outside funding. Across the top AI native companies, revenue per employee now averages several million. Small startups that are out earning per person some of the most established businesses ever built. And this has also triggered cuts to team sizes in the public markets. Meta, Microsoft, Block, and Coinbase have all recently announced significant layoffs as a result of studying this metric. So, it is a fair question to ask. If 15 people are making $80 million, where does the money go?

[00:07:12]Well, it goes to compute. These companies still have big costs. They're just a different kind. API calls, tokens, model training. The fundamental units of a startup cost has moved from salaries to compute. And that distinction matters more than it sounds.

[00:07:30]Payroll is a promise. Severance, not experience, morale, months to unwind.

[00:07:36]Compute is a dial. If the market turns, you scale it down before lunch. So the new breed is not just leaner. It carries a different kind of risk. A 400 person company in a downturn has to break promises to shrink. A 30% company just turns the dial. Investors noticed. And that is where the third domino falls.

[00:07:58]The one that impacts venture capital itself. Step three, the money has nowhere to go. Venture capital was built on a simple loop. Raise a fund. Write big checks into companies that need money. Watch them spend it on armies of people. Mark up the value, raise a bigger fund. Blit scaling was not just a strategy for founders. It was the demand side of the whole business. That machine needed companies that consumed capital.

[00:08:28]But those new companies do not consume it. Gamma is profitable. Midjourney never took a dollar of funding. Lovable at $400 million in revenue with 146 people generates cash. What exactly is a growth fund? was supposed to do with a company like that. There is no $200 million round to lead because there's nothing [music] to spend $200 million on. The best businesses of this era need investors much less than any generation of startups in history. So, where did all the money go? Well, I've done a video about this before, and well, nearly 75% of the money is going to a handful of AI labs, the ones training frontier models, the only businesses left that can genuinely absorb billions.

[00:09:16]They are swallowing nearly all of the venture investments on the planet. The capital that used to fund hiring sprees now funds GPUs. The deployment problem did not shrink. It just got redirected into the only furnace big enough to burn it. And for founders, there is an enormous consequence. A company that does not need capital does not sell it.

[00:09:40]Less money raised means less dilution, smaller preference stacks, founders and early teams keeping dramatically more of what they built. The best companies of this era may be the ones investors own the least of. And I'll be honest, for someone like me, this is great news. If you invest at seed before the company is proven, the new physics works entirely in your favor. We're watching it happen inside our own portfolio at Lobster Capital. Companies reaching profitability earlier, raising later and far more strategically, treating a round as a tool instead of a lifeline, which means that our early ownership does not get shredded through five rounds of dilution on the way up. and the growth underneath feels at the moment very sustainable, built on revenue, not on the next check or funding round. The people this era punishes are the mid-stage and midsize funds with capital but no place to put it. The people it rewards are whoever got in first and owns a piece of a machine that never needs rescuing again. Which brings us to the last domino, the one that changes how I read all startup pitches. Now, step four, the signal flipped. For 20 years, team growth was a key performance indicator. Funds literally reported portfolio headcount to their backers as evidence of progress. The question in board meetings was why haven't you hired faster? That question has inverted. It is now, why did you hire? Because think about what a 400 person org chart tells an investors in 2026. Best case, it's a company built before the shift carrying a structure from the old physics. Worst case, it is a signal that the product cannot do what the lean cohorts products do. It's a signal that this business still needs human routers, still pays the coordination tax, still converts capital into salaries instead of output.

[00:11:44]Either way, every one of those salaries is a cost the 15% competitor does not have. The org chart used to be the moat, but now it is the drag. Signals rarely reverse polarity. When they do, everyone still reading the old signal is pattern matching in exactly the wrong direction.

[00:12:03]Founders patting teams to look serious.

[00:12:05]Investors backing scale that is actually not sustainable. and even job seekers treating the fastest hiring companies as the safest bet. The whole industry spent two decades learning to read a gouge that now runs backwards. You can watch the inflection happen in an 18month window. April 2025, the Shopify memo, prove AI cannot do it before you hire.

[00:12:30]Clara CEO boasting his AI did the work of 700 customer service agents and letting the company shrink by attrition on purpose. Then lovable hitting 400 million with 146 people. And now Gartner forecasts that by 2030 80% of large organizations will have shrunk their engineering teams into small AI augmented ones. Remember Sam Almet in 2025 saying his tech CEO group chat runs a betting pool on the first year aperson company hits a billion dollar valuation.

[00:13:04]Less than 12 months later, the New York Times found the guy. Matthew Gallagher launched a weight loss telealth company called Medvi from his apartment with $20,000 and zero employees. AI wrote the code, made the ads, run the customer service. First full year, $41 million in revenue and $65 million in profit. His entire team is him [music] and his brother. Alman asked about it, said it looked like he had won the bet and that he would like to meet the guy.

[00:13:33]Blitzcaling assumptions held for a generation, but came apart in basically less than a year. Now, before you fire your entire team, here's the honest part. Clara did a U-turn. The loudest [music] AI replaces human story in the world quietly started rehiring people throughout this year after customer satisfaction fell apart on complicated cases. The AI handled routine volume brilliantly and failed where judgment mattered. So, ironically, Clara is now the case study every board sites when someone proposes deleting a department.

[00:14:09]Even the lean champions are staffing up.

[00:14:12]Cursor the AI coding company run with around 20 people to its first 100 million in revenue and then they grew to roughly 300 people the moment big enterprise customers arrive. selling to the Fortune 500 still takes [music] humans. So, the honest claim is 10 times fewer people. Nobody is seriously claiming zero. The productivity evidence is messier than the hype. One randomized trial found experienced developers were actually 19% slower using AI tools while believing they were faster.

[00:14:45]Self-reported gains are doing a lot of work in this [music] story. And tiny teams are not new. Instagram had 13 people when it sold to Facebook for a billion dollars. That was in 2012, years before any of this. Same thing with WhatsApp that had only 55 people and sold at 19 billion. Even Medvie, the twoman wanderer from earlier, is less a new species than an old one turbocharged. A middleman with great ads renting everyone else's infrastructure in a market where one single letter from a regulator could end it [music] all.

[00:15:20]Part of the revenue per employee spectacle is what early winners in new markets have always looked like. And if competition floods every niche, those numbers could compress hard. So the claim worth keeping is narrower and sharper. [music] Lean did not become mandatory. It became possible. But once it is possible, everyone competing against it has a problem whether they shrink or not. So let's go back to those two companies from the start and the question I left open. How do you really read team size now? Well, the important ratio is revenue per person, and it is becoming the first thing sophisticated investors check ahead of growth rate, ahead of total headcount, ahead of the logo slide. A company at $3 million per head is a very different machine than a company at $300,000 per employee. For 20 years, the industry asked founders how fast they could grow the team. The question that replaces it is harder and better. What is your revenue per person and why is it not higher? The companies with a great answer are the ones nobody can catch. The companies still bragging about headcount are the pitch you should worry about. Thank you [music] for watching. Don't forget to subscribe and I'll see you in the next one.