In March 2026, Daniel Nadler stood on stage at Nvidia's GTC conference and said something most people weren't ready to hear.
Nadler is the founder and CEO of OpenEvidence, an AI-powered clinical decision support platform valued at $12 billion. His company has fewer than 100 employees. Those employees, working through doctors who use OpenEvidence, reach approximately 300 million Americans every year.
He said: the world isn't prepared for what's coming.
He's right. I've been thinking about it ever since.
This Isn't a Story About Cutting People
The "sub-100 company" isn't a new way to dress up mass layoffs. It's a structural shift in what a company looks like when AI handles execution.
Block cut 40% of its workforce due to AI productivity gains. Jensen Huang told Nvidia's workforce to expect roughly 100 AI agents working alongside every human employee. AI agent stacks now cost between $300 and $500 per month and replace functions previously requiring $80,000 to $120,000 per month in payroll.
This is happening now.
36.3% of new ventures in 2026 are solo-founded, a figure climbing as AI agent reliability improves. Small founding teams with aggressive agent stacks are hitting product-market fit without building headcount to do it. One developer writing prompts and shipping code through a well-configured agent system now equates to what used to take an entire sprint team.
The 100-person company isn't dead because companies are struggling. It's dead because companies don't need 100 people to do the work of 100 people anymore.
And here's the part everyone's ignoring: this is good news. Not for shareholders, not for consultants. For HR. For culture. For anyone who has spent a career trying to make workplaces worth showing up to.
Why Culture Gets Harder, Not Easier
A smaller team doesn't mean a simpler culture. It means a more concentrated one.
In a 500-person company, a single disconnected manager affects perhaps a dozen people directly. The dysfunction diffuses. There are layers, there are handoffs, there are other teams absorbing the noise. In a 50-person company, one bad leader poisons the entire organisation. There's no buffer. There's no HR department large enough to contain the damage. There's nowhere for dysfunction to hide.
If your 50-person company has a leader who doesn't listen, gives no honest feedback, and rewards "always on" behaviour over results... everyone feels it. Within weeks.
I've watched this happen. You have too. A 10-person startup with a controlling founder doesn't need a toxic culture survey to tell people it's broken. They already know. The question is whether anyone feels safe enough to say so.
The smaller the team, the more each person's behaviour radiates outward. And the harder it is to get leadership wrong without paying for it immediately.
The Psychological Safety Gap Is Already a Crisis
Look at the data.
83% of executives surveyed say a company culture prioritising psychological safety measurably improves the success of AI initiatives. Nearly everyone in leadership agrees it matters.
Only 39% believe their organisations maintain a high level of psychological safety.
This gap isn't closing. It's widening as AI adoption accelerates and teams shrink.
Think about what this means in practice. Your team is experimenting with AI tools. Some of them are doing it in genuinely useful ways you don't know about. 32% of employees using generative AI at work hide it from their employers. They're not hiding it because they're lazy or secretive. They're hiding it because they don't feel safe sharing it.
When leaders emphasise AI primarily in terms of efficiency and cost reduction without equally addressing reskilling and future roles, they increase the perceived personal risk of being honest. People make calculations. "If I tell my manager I've been using AI to do my job in half the time, do they cut my role? Do they think I've been slacking? Do they think the AI is replaceable for me?"
So they stay quiet.
In a 200-person company, you lose some signal. In a 30-person company, you lose everything. Your competitive advantage... gone. Not to a competitor with better technology. To your own culture of silence.
What Leadership Looks Like at 50 People
The leader of a small AI-augmented team is not a manager in the traditional sense. They're not coordinating task handoffs or tracking hours. Agents handle execution. Humans handle judgment.
The leader's primary job becomes making sure the humans are making good judgments. Which means:
Make it safe to say "I'm not sure this is working." When an AI agent produces output nobody questions, errors compound quietly. The agent doesn't know it's wrong. Your people do. But only if they feel safe enough to raise it.
Make it safe to say "I think we should do this differently." Small teams move fast. This speed is a strength until it isn't. The only check on a fast-moving small team is a culture where people challenge direction without fear.
Make it safe to admit uncertainty. When your engineer doesn't understand what the AI model is doing, they need to say so. Not to look ignorant, but because nobody else is watching. In a large org, someone else checks. In a small one, you need everyone checking.
Make it safe to surface problems early. In a team of eight people, a problem ignored for two weeks doesn't stay contained. It becomes everyone's problem. The faster your team surfaces issues, the faster you fix them and move on.
None of this is complicated. Most of it is common sense. And most companies are terrible at it.
Leadership at 50 people is not coordination. It's the creation of conditions where your people feel safe enough to do their best work, flag their concerns, and tell you what the agent got wrong.
The Forcing Function Nobody Asked For
Here's what Nadler and Huang are getting right, and what most commentary misses.
The era of large headcounts hiding poor leadership is ending. You used to build a mediocre culture at scale and get away with it because volume created its own momentum. Volume absorbed poor management. Volume diffused bad decisions. Volume gave everyone enough to do so nobody had time to notice the rot.
Not anymore.
In a sub-100-person company, every hire is a culture decision. Every leadership behaviour is amplified. Every piece of feedback matters more. Every broken trust costs more.
MIT Sloan research argues sustainable differentiation doesn't come from technology. It comes from human creativity, anticipating customer needs, and building trust over time. The model you're using becomes a commodity. The culture you've built doesn't.
If you get this right in a 50-person company with a strong AI agent stack, you're building something no large, slow-moving competitor with 500 humans and a broken culture is able to replicate quickly. Their headcount is now a liability. Their bureaucracy slows adoption. Their poor feedback culture means their agents keep getting bad inputs and nobody says anything.
Your advantage is your people. Specifically, what your people feel safe enough to say.
What You Should Be Doing Now
This isn't a distant future problem. Sub-100 companies are operating at scale right now. If your team is growing, the time to build psychological safety is before you need it, not after you've realised it's gone.
A few things worth asking yourself:
- When someone on your team uses AI in a way you didn't expect, is your first reaction curiosity or suspicion?
- When a project isn't going well, do people tell you early or do they wait until it's unfixable?
- When your team disagrees with a decision, do they say so or do they comply and resent?
The answers tell you more about your culture than any survey.
The 100-person company is dead. The 50-person company with great culture and the right agents is what replaces it. Start building the culture now, while you still have room to get it wrong safely.
What's your plan for building psychological safety as your headcount shrinks and your agent count grows? I'd like to know what you're seeing.