Hone raises $60M for AI agents built to keep working

Hone is building enterprise AI agents meant to stay responsible for a business result, rather than disappear after answering a prompt. A company can assign one of its “Engines” a goal such as improving lead conversion, lowering vendor spending or reducing credit losses. The San Francisco startup announced a $60 million seed round from Benchmark and Index Ventures to develop the system.

The workflow starts with a team defining success, setting boundaries and connecting the Engine to relevant systems, people and processes. Hone says the software can rehearse against old tickets, deals and decisions before going live, allowing employees to correct its proposed behavior. Once deployed, it works through Slack, Microsoft Teams and email while retaining context between interactions.

Hone presents different Engines for specific operating jobs. One enriches and routes inbound leads before following up with prospects; another reviews software spending, collects missing context from employees and coordinates contract consolidation or renegotiation. Other examples cover credit risk, product adoption and customer commitments.

The company’s site displays customer testimonials from teams at Modal, Lucanet and Cognition, and says Engines have been deployed in revenue, procurement, risk, product and operations. Those references show the intended buyer is an organization handing over a continuing workflow, not an individual looking for a writing assistant. Hone hasn’t published pricing or quantified customer outcomes.

That autonomy also creates the central tension. A persistent agent needs access to company information and permission to act on it. Hone says customers can require approval for sensitive actions, audit every action, keep execution sandboxed and prevent their data from being used to train models.

Hone says replaying proposed changes across 125 internal scenarios reduced regressions from 33% to 5%. The result comes from the company’s own testing, but it makes the product’s challenge concrete: an agent operating over time has to improve without silently breaking behavior its customer already relies on.