The shift
From autocomplete to autonomy
For most of the last two years, AI in software meant an assistant in the editor that waited for an engineer to type and offered the next few lines. It was useful and is now widely adopted, but it left the shape of the work alone, because the engineer still owned every step.
Factory starts from a different premise. Its Droids take the parts of engineering that don’t need human judgment: testing, debugging, refactoring, migrations, code review and documentation. An engineer hands off an objective, and the Droid works across the codebase and comes back with a pull request for a person to approve. Engineers move to the outer loop of requirements, architecture and customers, while agents handle the inner loop of writing, testing and building.
The product
What Factory actually built
Factory runs where engineering already happens: the command line, Slack, GitHub and project tools like Linear. Droids pull context from systems like GitHub and Sentry, so they work from the same memory the team has.
For work too big for one agent session, Factory’s Missions split a large task into parallel tracks worked by multiple agents. That structure lets complex work run for hours or days without drifting, which is where most agents fail on real codebases.
Droid Computers give each agent a remote, persistent machine, so long-running work has somewhere to live. The result looks less like a coding tool and more like an operating layer for an engineering organization.
The moat
The moat is the environment the agent runs in
Factory’s founders argue that the hard part is the environment an agent runs in. CEO Matan Grinberg compares a coding agent to a Ferrari dropped into 1850s San Francisco, when the roads were still dirt. In an enterprise, the roads are documentation, test coverage, CI/CD, observability and the knowledge that lives in senior engineers’ heads.
The enemy of AI agents is tacit knowledge.Matan Grinberg, CEO, Factory, in an interview with McKinsey
Factory’s moat comes from aiming at that problem. It instruments the whole lifecycle so the system learns from every session, review and resolved incident, and that learning compounds inside the customer’s own walls. It was designed for sovereignty from the start, with hosted, bring-your-own-key, self-hosted and fully air-gapped options plus SOC 2 and ISO 42001 compliance, which regulated buyers in finance, healthcare and government require.
It’s also model-agnostic. Factory Router sends each task to the model that fits, which the company says cuts token spend by more than 60 percent, and customers can bring their own keys or run open models for sensitive code. Nobody is locked to one lab, and that orchestration is harder to copy than any single agent.
Hosted, bring-your-own-key, self-hosted and air-gapped deployments, with SOC 2 and ISO 42001 compliance.
Timing
Why this matters now
In April 2026 Factory raised $150 million at a $1.5 billion valuation in a round led by Khosla Ventures, and said revenue had doubled month over month for six straight months. In September it announced another $200 million at a $5 billion valuation, with Blackstone, Khosla, Sequoia, Insight and NEA among the backers. Hundreds of thousands of developers use Droids at enterprises including Nvidia, Adobe, Palo Alto Networks, RBC and T-Mobile.
These are some of the most demanding engineering organizations in the world, well past the pilot stage, which tells you the product holds up against real codebases and real compliance teams. Because Factory’s system improves the longer it runs in your environment, the teams that start earlier build a lead that compounds.
