Instinct just raised a billion dollars for a product I can’t test, and neither can you.
That’s the whole story, and it’s also the whole problem. On Monday the company confirmed a $1 billion Series C at a $10 billion valuation, with Sequoia Capital, Benchmark Capital, and Coatue on the cap table. The previous round valued the company at $2.5 billion. Meanwhile, Instinct is still in early access, running the invite-only test it launched in August 2026.
Four times the valuation. Same closed door.
What we actually know about the product
Instinct is building a personal agent that performs tasks on your behalf without supervision. The headline capability is a concierge function: it makes bookings and places phone calls. That’s the pitch. Restaurant reservations, appointments, the kind of small friction that eats twenty minutes of your week.
I want to be fair here, because that’s a genuinely useful thing to automate. Phone calls are the last unautomated interface in most people’s lives. Businesses that won’t take an online booking, clinics with hold music, the contractor who only answers between 8 and 9 a.m. An agent that reliably handles that is worth paying for.
The operative word is reliably. And that is exactly the part nobody outside the invite list can evaluate.
Why I’m not impressed by the number
A $10 billion valuation on an invite-only early access product tells you about investor conviction. It tells you nothing about whether the agent completes a booking correctly on the first try, what happens when the restaurant’s host misunderstands it, or how it behaves when a call goes off-script.
Those are the only questions that matter for an autonomous agent. Everything I review on this site lives or dies on failure modes, not demos. A concierge agent that works 90% of the time isn’t 90% useful — it’s a thing you have to double-check every time, which means you’ve added a step instead of removing one. The gap between “impressive demo” and “I stopped verifying its work” is where most agent products quietly die.
The invite-only problem
Closed betas are a legitimate engineering choice. You cap load, you watch logs, you fix things before a million people hit them. Fine.
They’re also a marketing instrument. Scarcity generates buzz, buzz generates coverage, coverage generates term sheets. Instinct’s fundraising was reported by The Information before the company confirmed it, which is a familiar sequence: the round leaks, the narrative builds, the press release lands. None of that is wrongdoing. But it means the loudest signal around this company right now is attention, not verified performance.
And attention is the cheapest thing in AI. I’ve watched agent startups generate enormous noise on a demo video and then ship something that can’t hold a session together for four minutes.
What would actually change my mind
I’m not writing Instinct off. Sequoia, Benchmark, and Coatue don’t all show up for nothing, and a billion dollars buys a lot of runway to solve hard problems. Autonomous phone calls are a hard problem. If they’ve cracked it, that’s real.
Here’s what I’d need to see before recommending it to anyone:
- Open access, or at minimum a published success rate on completed bookings and calls
- Clear behavior when the agent is uncertain — does it stop and ask, or does it improvise?
- What it does with your calendar, contacts, and payment details, and who can see them
- Whether the person on the other end of the call knows they’re talking to software
- Recovery behavior when a task half-completes, because half-booked is worse than not booked
That last one deserves emphasis. Autonomous agents fail in the middle. A chatbot that gives a bad answer wastes ten seconds. An agent that books the wrong night, or books twice, or hangs up on a receptionist mid-sentence, creates work in the real world that someone has to undo.
My read
Instinct has a big round, a big valuation, and a product category worth chasing. It does not yet have public evidence, and a $10 billion price tag isn’t evidence. The four-fold jump from $2.5 billion happened without the rest of us gaining a single new data point about how well the thing works.
If you got an invite, use it, and tell me what breaks. If you didn’t, there’s nothing here to act on yet — just a valuation and a promise, in that order.
I’ll review it properly the day I can actually run it. Until then, the funding round is news about investors, not news about agents.
🕒 Published: