Hone AI Agents Startup Lands $60M Seed From Benchmark, Index

A large first check for a five-month-old company

Hone, a San Francisco startup founded by former employees of OpenAI and Cognition, has raised a $60 million seed round led by Benchmark and Index Ventures 135. The company is about five months old 25. Most outlets put its valuation at $285 million and attribute that figure to Hone itself 135. One tracker is more cautious. It says the company's own launch announcement confirms the round and its leads but does not state the valuation, so it treats the $285 million as a figure reported by Bloomberg rather than company-filed data 2. The difference is small, but readers should hold the valuation a little more loosely than the round size.

Other participants include Elad Gil, Hanabi, Definition, Diffusion, Lux, SV Angel and Align 4. Benchmark's Peter Fenton and Index's Shardul Shah are joining the board 134. Board seats from two lead partners at the seed stage are a strong signal that both firms expect this to become a substantial company.

What Hone says it is building

Hone's pitch rests on duration and ownership. It is building what it calls "Engines," persistent AI agents meant to own a business outcome over weeks or months instead of waiting for prompts or finishing isolated tasks 24. CEO Moritz Stephan described the goal as "building AI that owns outcomes." In his framing, the system studies a team's goals and responsibilities, deploys engines to take them on, and checks in with humans when needed 5.

The examples so far are concrete and modest. They include evaluating sales leads and acting as a procurement coordinator 13. Bloomberg draws the distinction this way: many vendors build software that helps a sales team judge leads, while Hone wants an engine that does that work on its own and proactively 5. The company also calls itself an applied research lab working to close the gap between what frontier models can do and the economic results they actually produce inside organizations 4. Fenton said Hone is developing cybersecurity guardrails for its agents 1. That matters for any system expected to act unsupervised for long stretches.

The Cognition loop

The most notable feature of this deal is how closely Hone is tied to Cognition, the company behind the Devin coding agent 3. Stephan was Cognition's chief of staff and previously did agent research at Stanford 13. Cognition is also one of Hone's early customers, with the work centered on go-to-market operations 1. Cognition CEO Scott Wu has invested personally 1. AI inference startup Modal is the other named early customer 1.

This arrangement cuts both ways. A friendly early customer gives a young agent company something it badly needs: real workflows, real data and tolerance for rough edges while the product matures. Having a well-known AI company running Hone's engines in its own sales operations is also a credible early proof point. But when a former employer is also a customer and its CEO is an investor, the first deployment is not quite an arm's-length test. For now, "Cognition uses it" partly reflects relationships and only partly reflects product-market fit. The more telling evidence will come from customers with no personal ties to the founding team.

Why it matters

Bloomberg places Hone in a crowded market of companies selling AI software that automates complex work, especially in coding and customer service 5. Hone's bet is that the next step is broader. It wants agents that take on ongoing operational responsibilities, not just help with discrete tasks 5. If that works, the unit of value moves from "tool a worker uses" to something nearer "staffer a team assigns," which Bloomberg's framing of agents as professional staffers makes explicit 5.

That ambition also explains the size of the round. Agents that run for weeks need dependable memory, the judgment to know when to escalate to a human, and security controls strong enough that a company will let software act on its behalf. Each of these is hard. The $60 million suggests investors expect a long, costly research effort before broad deployment, which fits Hone's description of itself as a research lab 4.

The reading

The facts are not in dispute: a $60 million seed, two top-tier leads, and a team with strong lineage. What remains unproven is whether "owning outcomes" is a real product category or a rebranding of agent automation that rivals are already chasing. Hone's early traction runs through its founders' own network, and the most precise valuation detail depends on whose announcement you read. The concept is promising and the backers are serious. The independent evidence is still to come.