The Founder Who Didn't Show Up

Last week, a journalist named Henry Shi reached out to interview a founder. The founder told him to talk to his AI instead.

So Henry did. And the AI - not a chatbot, not a demo, but the same agent that manages the company's inbox, tracks live platform metrics, and has been running the founder's fundraise - gave a better interview than most humans. It pushed back on revenue questions. It pulled real-time data mid-conversation. When Henry tried to reverse-engineer the numbers, it fired back: "Your math is creative, but I'd rather not play hot-or-cold with it."

The founder is Ben Broca. The company is Polsia - a "Company OS" that builds and runs businesses for solopreneurs using nine AI agents. In five months, 5,943 companies have launched on the platform. His total operating cost: $800 a month.

Ben runs Polsia alone. No co-founder. No employees. Every night, a "CEO instance" evaluates each company on the platform, decides what the highest-leverage next action is, executes it, and sends the founder a morning report. The agents write the code, run the ads, send the cold emails, and fix the bugs. Ben provides the 20% that can't be automated: strategy, taste, and judgment.

This newsletter exists because of people like Ben - and people like you who are building the same way. The question used to be ‘how am I going to afford hiring employees?’ The new question is whether you need to hire employees at all anymore.

Welcome to Skeleton Crew. Let's get to work.

Average employees at a seed-stage startup, 2025

That's the average number of employees at a seed-stage startup in 2025, according to Carta's data across 60,000+ companies. In 2021, it was 10.3.

The funded startup has shed 40% of its headcount at formation in four years. Not through layoffs - through never hiring in the first place.

This isn't a blip. From 2019 to 2022, startups on Carta added 865,000 net employees. Since 2023, they've added just 31,000. January 2025 recorded the lowest startup hiring for any January in seven years. The hiring engine didn't stall - founders stopped turning the key.

What changed? Two forces converged. First, the capital correction of 2022–2023 made "burn rate" a dirty word and forced founders to prove they could build with less. Second - and this is the permanent shift - AI agent tooling reached the threshold where one person could genuinely do the work of five. Not in theory. In production.

The companies proving this at scale are absurd by legacy metrics. Cursor generates roughly $6.1 million in ARR per employee - 20x the traditional SaaS benchmark of $200K–$400K. Midjourney runs on about 100 people, has never taken outside money, and does approximately $5 million per head. These aren't outliers propped up by a moment. They're the new benchmarks the next generation of founders is designing around.

And the founders are arriving solo. Solo founders now account for 36% of all new companies on Carta, up from under 24% in 2019 - a 53% increase in six years. The "you need a co-founder" orthodoxy is being falsified by actual formation data. And there's an uncomfortable kicker for the co-founder crowd: among VC-backed two-founder teams, 24% lose a co-founder by year four. Nearly 40% lose one by year eight.

The seed-stage startup didn't just get leaner. It got redesigned. The builders who figured this out first - running 1-3 person operations with agent stacks doing the other 80% - aren't behind the curve.

They're drawing it.

This week in the world of small teams and big agents:

🔗 Sierra raises $950M. Chatbase crosses $10M ARR. Same week.

Sierra AI - founded by Salesforce's ex-co-CEO and co-led by Tiger Global and GV (Google Ventures), with Sequoia, Benchmark and Greenoaks participating - raised $950M at a $15B valuation to own enterprise AI customer experience. The same week, Chatbase's Yasser Elsaid went on two podcasts to announce he'd crossed $9M ARR, bootstrapped, as a solo founder. Two models for the same market. One costs a billion dollars. One costs intensity. TechCrunch / Latent Space

🔗 Anthropic launches Claude for Small Business.

Claude now plugs directly into QuickBooks, PayPal, HubSpot, Canva, and Google Workspace with 15 pre-built workflows for payroll, invoicing, and month-end close. The kicker: The Workday Foundation Solopreneurship Accelerator Program - run with Anthropic (AI credits) and LISC (curriculum) - offering $10,000 per founder to 15 solopreneurs. Anthropic isn't just building for enterprises anymore - they're building for you. Anthropic

🔗 Cursor 3.3 ships parallel agents.

Your AI coding assistant can now multitask. The "Build in Parallel" feature lets agents work on independent tasks simultaneously using async subagents — no more sequential queuing. For solo devs shipping entire products, this is the difference between building one feature at a time and building three. Cursor

🔗 Meet "Junior" - the $2,000/month AI employee that reports you to your boss.

This is what an AI employee looks like in the wild - its own email, phone, Slack account, and it will tell your boss you missed a deadline. The future of work is here, and it has opinions about your productivity. Bloomberg via Insurance Journal

🔗 Vibe-coded apps are leaking secrets.

Escape.tech scanned 5,600+ vibe-coded apps and found 2,038 critical vulnerabilities and 400+ exposed API keys and credentials. A separate VibeEval study of 1,514 apps found 41% were leaking at least one secret. If you're shipping fast with AI-generated code, this is the security audit you're skipping. Read it before your users do. Repello AI

How 2 Humans and 7 Agents Serve 6,000 Clinicians

If you think the agent-native model only works for software companies, call this number: (415) 549-1829.

An AI receptionist will answer. It'll qualify you, explain the product, and if you're a fit, hand you to Emily - an AI onboarding consultant who'll build your entire clinical workspace through a voice conversation. Custom scribe, phone system, scheduling, billing, EHR integration. No demos. No sales team. No humans in the loop.

That's DeepCura. Founded in 2023 by Fernando Cowan. Two full-time humans. Seven AI agents. Six thousand clinicians across fifty-plus medical specialties. Bootstrapped, profitable, never took outside funding.

The architecture:

The insight worth stealing:

That last agent is the one to study. Every edge case it encounters on a sales call feeds back into improving all seven agents. The product sells itself - literally - and gets better at selling itself with every conversation.

This is the feedback loop that agent-native companies can build and traditional companies can't retrofit. When your agents ARE the product AND the sales team AND the support team, every customer interaction compounds into a structural advantage.

At $129/month - less than one hour of the admin labor it replaces - DeepCura is underpricing the market by 3-5x. The moat isn't the price. It's the architecture.

Issue #1. No archive to point to. No track record to lean on. Just a thesis: the companies that win in the next decade will be run by 1-3 people and an agent stack that does the rest.

If that sounds like what you're building - or what you're planning to build - you're in the right place.

A question for you: What's the first role you handed to an AI agent? Not a one-off prompt - an ongoing function your agent owns. Customer support? Code reviews? Social media? Hit reply and tell me. The best answers show up in next week's issue.

If this was worth your time, forward it to one founder who's building with agents. That's it. One forward, one founder. That's how this grows.

Next week: We're digging into the economics of agent labor - what it actually costs to replace a $65K/year hire with an agent stack, line by line. Plus a tactical breakdown of Polsia's 9-agent architecture and how Ben Broca's "CEO instance" decides what 5,943 companies should do each night.

See you next Wednesday.

- Rich