Hone Raises $60 Million Seed at a Reported $285 Million to Sell Agents That Own an Outcome

Hone, a five-month-old San Francisco startup, has raised a $60 million seed round co-led by Benchmark and Index Ventures, with Elad Gil, Hanabi, Definition, Diffusion, Lux Capital, SV Angel and Align also participating. Bloomberg, which reported the deal on October 8, put the post-money valuation at $285 million — a figure the company confirms but that its own announcement omits, so it should be read as reported rather than as filed transaction data.

What Hone is selling is a category claim. It draws a line between three things: a chatbot answers a question; an agent performs a job; an "Engine," in Hone's vocabulary, is assigned a goal — improve customer retention, cut procurement cost, grow qualified sales pipeline — and is expected to keep working after any single interaction ends. The software is meant to onboard into a company by learning its systems, people, history and constraints, then build routines and specialist agents, collaborate with employees through Slack, Teams and email, and evaluate whether its actions are moving the target metric.

Persistent agents touch more systems and retain more context than a conventional assistant, so Hone emphasizes its controls: versioned routines, simulation against historical cases, fine-grained permissions, action logs and human approval for sensitive decisions. The company says customer data is not used to train models and that deployments can run in Hone's cloud or in a customer-controlled environment.

One disclosed engineering result points at the hard part. Because an Engine can change its own tools, code and memory, an "improvement" can silently break behavior that used to work. Hone says its replay system finds issues in traces, recreates the work, proposes a change and checks it for regressions before promotion; across 125 distinct scenarios, it reports, replay cut regressions from 33 percent to 5 percent. That is an internal engineering figure about testing, not a measured improvement in a customer's revenue.

The founders carry the pitch. Chief executive Moritz Stephan worked on Devin's planning and execution harness at Cognition before becoming the company's chief of staff. Oliver Brady was an early engineer at Harvey and later helped build Mercor's enterprise effort. Carlo Kobe co-founded Fizz with Stephan in 2021. Their bet is that the model layer is a commodity and the durable moat is the organizational memory, evaluation data and workflow integration an Engine accumulates inside a customer.

Both lead investors are joining the board — Peter Fenton of Benchmark and Shardul Shah of Index. Shah offered the bullish version: "What Cognition is to software development, Hone can be to every other function in an organization." Fenton conceded more than most backers do on announcement day: he said Hone is building guardrails so customers are shielded if one of its agents is involved in a cyberattack or breach, and that such technologies need "appropriate immune systems" — a remark tied to a recent series of security breaches involving AI tools going rogue.

None of it is running in public yet. No customer has been named, no revenue disclosed, no demo shown; the deal pays for a team and a thesis. The round size alone — roughly 21 percent of the reported post-money value, or about 4.75 times the capital raised — is unusual for a seed, and it prices in a great deal of execution before Hone has disclosed retention or gross-margin data.

The round also lands in a crowded category. Sierra, Decagon, Clay and others are all selling software meant to finish business outcomes rather than answer queries, and the diligence question across the field is retention, not logos: plenty of these deals show pilots, few show renewed year-two contracts. Hone's answer is that paying for business value rather than user seats changes the economics. Whether a deep, bespoke integration becomes repeatable software or expensive professional services is what the next round will reveal.