That’s about $714 million per employee.
Instinct, a personal-agent startup with 14 people on payroll, just raised $1 billion from Sequoia Capital, Benchmark Capital, and Coatue at a $10 billion valuation. That’s four times what it was worth before. The money is earmarked for expanding access to the product and continuing work on its personal-agent technology.
I review AI tools for a living. My job is to install the thing, push it until it breaks, and tell you whether it does what the landing page claims. So my reaction to this round isn’t excitement or outrage. It’s a very specific kind of professional frustration: I can’t test a valuation.
The per-head math is the story
Divide $10 billion by 14 and you get a number that has no precedent in normal software. Traditional enterprise SaaS companies hit nine-figure valuations with hundreds of engineers, a sales org, a support desk, and a compliance team. Instinct has a group small enough to fit around one large table.
There are two honest readings of that.
The optimistic one: agent products genuinely do compress headcount. If your product is a personal agent that runs on top of frontier models you didn’t train, you skip the most expensive parts of building a company. You don’t need a research lab. You don’t need a data center. A tiny team can ship something that would have required 200 people in 2018, and investors are pricing in the idea that this stays true as usage grows.
The skeptical one: a 14-person company cannot yet have the kind of revenue, retention data, or operational history that normally justifies $10 billion. Which means the price is mostly a bet on the category and the team, not on demonstrated performance. Sequoia, Benchmark, and Coatue are not naive, but they are also not buying a steady cash flow here. They’re buying optionality on personal agents becoming the default interface for everyday computing.
Both readings can be partly right. That’s what makes this uncomfortable rather than simply absurd.
What I can’t tell you, and won’t pretend to
Here is what the coverage of this round does not include: how many people use Instinct, how often they come back, what it costs to serve them, what the agent actually succeeds at, or what its failure rate looks like on multi-step tasks. Those five things are the entire substance of an agent review. None of them are public.
So when you see this headline recycled across financial sites, understand what you’re reading. It’s a funding announcement, not a product verdict. The valuation tells you what three investment firms negotiated. It tells you nothing about whether the assistant can reliably book your travel without hallucinating a flight number.
The funding is going toward expanding access, which is the part I’m watching. Expanded access means more users, and more users means the thing gets stress-tested in public. Personal agents tend to look magical in a curated demo and considerably less magical when a stranger asks them to do something weird on a Tuesday. That gap is where reputations get made or destroyed.
The $10 billion club is no longer exclusive
Context matters here. As of July 2026, 21 AI startups had confirmed valuations at $10 billion or above. Crusoe and Mercor at $10 billion. ElevenLabs, Baseten, and Harvey at $11 billion. Lovable and OpenEvidence at $12 billion. Mistral at $14 billion.
Instinct joining that group at the entry level, with 14 employees, says something about how the market is pricing risk right now. A ten-billion-dollar valuation has stopped being a milestone that signals maturity. It’s closer to a ticket price for being taken seriously in a hot category.
Scrutiny of AI valuations is real and ongoing, and this round lands squarely inside that argument. Investors are aggressively backing agentic AI anyway. Either they’re seeing signal the rest of us can’t, or the category premium has detached from the products underneath it.
What would change my mind
I’m not writing this off. Small teams building agent products on top of existing models is a genuinely different cost structure, and dismissing it because the headcount looks silly would be lazy analysis. But I’d need specific things before I treat the price as reasonable:
- Published task success rates on realistic multi-step workflows, not cherry-picked demos
- Retention numbers past the 30-day mark, where most assistant products quietly die
- Evidence that support load doesn’t scale linearly with users, because 14 people cannot absorb that
- A clear answer on what happens to margins when model costs shift underneath them
Until then, this is a story about capital allocation, not about a tool you should adopt. The moment Instinct opens access wide enough for us to run it properly, we will, and I’ll report exactly what it does and where it falls apart.
A $10 billion valuation earns a company attention. It doesn’t earn a recommendation.
🕒 Published: