The next unicorns will have five employees.
Why AI agents that work while you sleep — not chatbots you prompt — are about to change who can build a billion-dollar company.
TL;DR: AI won't make you more productive if it still needs you to prompt it. The founders who win are building companies where AI runs operations proactively — while they sleep, eat, and think. We built that infrastructure for ourselves first, realized it was more valuable than our original product, and pivoted. This is that story.
You have 24 hours in a day. AI doesn't change that.
Every tool in your stack — ChatGPT, Claude, Cursor — is waiting for you to prompt it. The moment you close the tab, it stops. When you're sleeping, eating, or just living your life, your AI workforce is idle.
The bottleneck in 2025 isn't intelligence. Models are smart enough. The bottleneck is what engineers call the harness — the infrastructure, memory, and skills that surround a model and let it work without you. Without a harness, AI is a very smart tool that only works when you're holding it. With a harness, it's a co-worker.
We know this because we spent two years building exactly that.
From Basalt to Pancake
We started as Basalt, an AI observability and evaluation platform. Our days were spent embedded with some of the best AI engineering teams in San Francisco — learning an enormous amount, flying constantly between SF and Paris, and falling badly behind on running our own company.
So we used OpenClaw to put operations on autopilot. Agents handling outreach, content, research, internal coordination — all running in Slack, all working while we slept.
It worked better than expected. Then our customers started asking about the internal tooling more than the product we were selling. We were getting more excited demoing it than demoing Basalt.
The conclusion was obvious. We pivoted. Basalt became Pancake.
What we believe
Scaling used to mean hiring. More revenue required more people — salespeople, ops, support, coordination overhead. AI breaks that relationship.
The companies we find most exciting are built AI-first from day one: AI handles the operational layer, humans focus on what AI can't do. We think there are now two ways to run a company — AI-first, where AI runs the company and humans help, or human-first, where humans run the company and AI waits to be prompted. The second model is going to become increasingly hard to compete with.
We also believe in Jevons' paradox: as AI makes capability cheaper, more people will use it, not fewer. The founders who benefit most won't be large companies with existing leverage — they'll be the ones who couldn't previously afford to compete. The one-person company running like a ten-person team. The two-person startup entering a market that previously required a full GTM org.
We think the next decade will produce more small companies building significant things than any decade before it. We're building the infrastructure to make that possible.
What Pancake is
Pancake is the infrastructure that takes you from $1 to $1M in revenue without hiring. It's the harness we described above — memory, skills, schedules, coordination — packaged so any founder or small team can use it.
Here's what makes it different from the hundred other "AI agent" tools:
Guided autonomy, not autopilot. Pancake runs what you tell it to run and keeps you in the loop for the rest. You stay in control. It surfaces decisions that need human judgment and handles everything else proactively — across Slack, iMessage, and email.
Works solo or multiplayer. If you're building alone, you get the full AI workforce to yourself. If you have a team, every person gets the same context, the same infrastructure, the same leverage — no one is working with a different version of reality. Most tools force you to choose. Pancake scales with you.
Open source agents, built by the community. The agent skills are on GitHub. You can inspect them, fork them, contribute new ones. No black box. No mystery about what's running on your behalf.
Transparent everything. Agents are community-sourced. Tokens are passed through at public API prices. No credit system, no hidden markups. You see exactly what you're paying for and why.
And the strongest signal we can give you: Pancake runs on Pancake. Every piece of our own operations — outreach, content, engineering coordination, research — is run by the same agents you get access to. We're not selling something we wouldn't use ourselves. We're selling the thing we can't live without.
Frequently asked questions
- What is the harness problem in AI?
- The harness is the infrastructure, memory, and skills that surround a model and let it work without a human actively prompting it. Without a harness, AI only works when you're holding it. With one, it works around the clock — executing tasks, coordinating agents, and surfacing decisions while you focus on other things.
- What is an AI-first company?
- An AI-first company is one where AI handles the operational layer — growth, engineering coordination, ops — and humans focus exclusively on what AI can't do: relationships, judgment, creativity, and strategy. The company runs on AI infrastructure, not headcount.
- What is Pancake?
- Pancake is AI co-founder infrastructure that takes you from $1 to $1M in revenue without hiring. It runs what you tell it to run, keeps you in the loop for everything else, and works for your whole team across Slack, iMessage, and email. The agents are open source, community-built, and available on GitHub. Pancake runs on Pancake.
- Why did you pivot from Basalt to Pancake?
- We built the agent infrastructure to run Basalt, our observability platform, without falling behind on operations. It worked better than expected. Our customers started asking about the internal tooling more than the product we were selling. The conclusion was obvious — we were building the wrong thing.
