Last Updated on July 7, 2026 by Taya Ziv
A company came out of stealth this week and got handed a $500 million valuation for software that does insurance underwriting. Read that sentence one more time, because every word in it is doing work. Underwriting. The thing a person in a beige office has done with a spreadsheet and a checklist since before you were born. That is the business Founders Fund and Kleiner Perkins just decided was worth half a billion dollars.
If your first reaction is “wait, that’s the boring part of finance,” then good. You’re paying attention. Because the boring part is exactly the point, and it’s the most useful signal a pre-seed founder got all month.
What actually happened
On June 11, a San Francisco startup called Poetic stepped out of stealth with a $50 million Series A at a $500 million valuation. The round was co-led by Founders Fund and Kleiner Perkins, two firms that do not write checks for fun, with an angel check from Mira Murati, who used to run things at OpenAI. The product is not a chatbot. It’s a set of AI agents that chew through insurance underwriting and lending paperwork, the document-heavy, rule-heavy grind that armies of analysts currently do by hand.
Now here’s the part that turns one deal into a pattern. Look at who else raised that same week. Jedify, in New York, pulled $24 million to build “context graphs” so enterprise AI agents stop hallucinating on company data. Capsa AI raised $18 million to organize the deal documents and spreadsheets that private equity firms drown in. nesto, up in Montreal, took roughly $230 million to run AI mortgage origination. A Tel Aviv company, Aryon, got $29 million to catch cloud misconfigurations before they ever hit production.
None of those will trend on tech Twitter. Not one of them is building artificial general anything. They each picked a single dull, expensive, regulated job that a human does today, and they’re automating it end to end. And the smart money lined up around the block.
Why this is the signal, not the noise
Here’s the trap I watch founders fall into every single week. The headlines this month were Jeff Bezos raising $12 billion for Prometheus and a German robot company raising $1.4 billion. So a founder reads that, gets a little drunk on it, and concludes the game is “go big, go frontier, raise a fortune.” That’s the wrong lesson, and worse, it’s a lesson you literally cannot act on. You don’t have Bezos money or a robotics supply chain sitting in a warehouse.
The deals you can actually learn from are the $18 to $50 million ones, because those are companies that looked exactly like yours eighteen months ago. Small team, one idea, no product anyone could see yet. And every single one of them won by being narrow and boring on purpose.
Think about why boring works right now. Two years ago, having an AI feature was a moat. Today the model is a commodity, it costs almost nothing, and your neighbor’s teenager can call the same API you can. When capability is basically free, the only thing left worth paying for is the stuff that’s hard to copy. And it turns out the hard-to-copy stuff isn’t intelligence anymore. It’s knowing exactly how a mortgage gets approved across a dozen states, which three fields a PE associate actually checks at 2am, what makes a seasoned underwriter quietly say no. That’s domain knowledge, regulatory scar tissue, and trust. You can’t prompt your way to it. It’s the same logic that just pushed 55 billion dollars into robots because atoms are genuinely hard to copy, except here the moat is process knowledge instead of hardware.
Boring has a quieter advantage too: almost nobody wants to do it. Every clever founder in your cohort is building the same AI notetaker, the same email assistant, the same “second brain.” Meanwhile the unglamorous jobs just sit there, worth billions, with a tiny line of competitors, because the work sounds like a yawn at a dinner party. This isn’t new, by the way. The last batch of Y Combinator was quietly dominated by startups doing deeply boring things, and that was not an accident.
The take, and where I might be talking my own book
So here’s my honest read. The AI gold rush just entered the phase where the money stops paying for “smart” and starts paying for “does one specific painful job that someone will actually write a check for.” Underwriting. Claims. Compliance. Mortgage ops. The back office everyone ignored because it wasn’t sexy is now the single best place for a small team to build something defensible.
But let me argue against myself for a second, because I could be pattern-matching on one good week and selling it to you as a law of nature. A $500 million valuation on a company that’s still in stealth is not proof the product works. It’s proof that two famous funds and an OpenAI alum believe it will, which is a much cheaper and much more common thing. We have watched a lot of marquee-backed AI startups get fat valuations and then quietly vanish. By one count, only around 130 of the thousands of AI-agent companies out there are doing anything real. Poetic could be the next category winner or the next cautionary tale. I don’t know. If I’m being straight with you, neither do its investors.
What I’m more sure about is the direction, not any single name. When boring, regulated, back-office work starts pulling Founders Fund money in the same week from five different angles, that’s not a fluke. That’s capital telling you where it thinks the defensible value moved. You can argue with one deal. It’s a lot harder to argue with the pattern sitting underneath it.
What to build instead
Stop trying to build something impressive and start trying to build something specific. Go find a job that is boring, expensive, done by humans today, and ideally wrapped in enough rules that nobody fun wants to touch it. Insurance, lending, compliance, logistics, healthcare billing, government paperwork. The uglier it sounds, the better.
Then go painfully narrow. Don’t build “AI for finance.” Build the one thing that takes a document a loan officer hates and turns four hours into four minutes. Own that single workflow so completely that ripping you out would cost the customer a month of pain. And sell it before you’ve built the whole thing, the way you should have been doing anyway. The model is the easy part now, which by definition means it’s the part worth the least. The hard part, the part worth the $500 million, is knowing the boring job better than the people who do it for a living.
The most exciting companies of 2026 are going to look incredibly dull from the outside. Go be one of them.
