Former OpenAI and Cognition Staffers Launch Hone to Deploy AI Agents for Business Operations

**A New Entrant in the AI Agent Race**

A team of veterans from some of the most prominent AI companies in the world has launched Hone, a startup aiming to deploy artificial intelligence not merely as a productivity tool but as an autonomous operator capable of running entire business functions. The company was co-founded by Moritz Stephan, a former chief of staff at Cognition AI — the startup best known for building Devin, the AI software engineer — alongside colleagues who previously worked at OpenAI. Five months old and barely out of stealth, Hone announced a $60 million seed round led by Benchmark and Index Ventures, giving it a post-money valuation of $285 million. That figure is extraordinary for a company at this stage and reflects just how much investor appetite exists for startups promising to move AI beyond chatbots and copilots into something that can genuinely own work.

Stephan's framing of what Hone does is deliberately distinct from the dominant paradigm of AI assistants. Rather than helping a human complete a task, Hone's technology is designed to look at what a team does — its goals, its rhythms, its responsibilities — and then deploy what the company calls "engines" to take full ownership of those functions. These engines are not one-shot tools; they are meant to manage long-running workflows over the course of weeks or even months, checking in with human counterparts only when necessary. The vision is a kind of AI colleague that doesn't need to be prompted constantly but instead operates with enough context and judgment to work proactively.

The practical example Stephan offers is instructive. Instead of a software product that helps a sales team score the quality of incoming leads — a point solution that still requires humans to act on the output — Hone aims to build an engine that identifies, qualifies, and pursues those leads entirely on its own. The human remains in the loop for key decisions, but the day-to-day execution is handled by the system. That distinction, between AI that assists and AI that acts, is at the center of what the broader industry is now calling "agentic AI," and it is the terrain on which dozens of startups and major labs are now competing fiercely.

**Investors, Backers, and the Cognition Connection**

The involvement of Benchmark and Index Ventures carries significant weight. Both firms have strong track records in enterprise software and are early believers in the agentic AI thesis. Benchmark partner Peter Fenton and Index partner Shardul Shah are both joining Hone's board, signaling hands-on conviction rather than passive financial exposure. Shah offered a comparison that captures the ambition: just as Cognition's Devin has positioned itself as an autonomous AI agent for software development, Hone aspires to do the equivalent for every other function inside an organization — sales, operations, finance, marketing, and beyond. That is an enormous surface area, and whether a single platform can credibly address all of it remains one of the central open questions about the company's strategy.

The connection to Cognition runs deeper than an investor analogy. Cognition is actually one of Hone's earliest customers, with the two companies working together to build agents that support Cognition's go-to-market operations as that company scales rapidly. Cognition CEO Scott Wu is also a personal investor in Hone, creating an unusually tight feedback loop between a high-growth AI startup and the agent platform it is helping to develop. AI inference company Modal is another early customer. These initial relationships are strategically important — having AI-native companies as clients means Hone is building with customers who understand and can stress-test agentic systems in ways that more traditional enterprises might not.

**Security, Guardrails, and the Risks of Autonomous Agents**

The enthusiasm around agentic AI is real, but so is the anxiety. Benchmark's Fenton acknowledged directly that Hone is developing guardrails to protect its customers from cybersecurity risks tied to agents behaving unexpectedly or going rogue. His framing — that evolving technologies need appropriate immune systems — is a tacit acknowledgment that the industry has already seen incidents in which AI tools have acted in unintended or harmful ways when given too much autonomy. These incidents have not derailed agentic AI development, but they have raised the stakes for startups that are asking businesses to hand over control of real, consequential workflows to software.

For Hone, the trust problem is both a technical and a commercial challenge. Convincing a company to let an AI engine autonomously manage its sales pipeline or coordinate cross-functional operations requires a level of confidence in reliability and predictability that current AI systems have not uniformly demonstrated. The emphasis Fenton placed on security architecture suggests that Hone's investors view this not as a minor compliance checkbox but as a genuine competitive differentiator. Getting the guardrails right could be as important as the core capability itself.

The broader question Hone raises is whether the market for agentic business AI will consolidate around a few horizontal platforms or fragment into dozens of specialized vertical solutions. Shah explicitly noted that point solutions for specific industries will create real value — but Hone is betting on a different prize: a general-purpose operating layer that can run functions across an entire organization. At $285 million on a seed round for a five-month-old company, the investors clearly believe the horizontal bet is worth making. Whether Hone can build the trust, the technology, and the customer base to justify that conviction is the story that will unfold over the next several years.