Factory raises $200 million for AI coding agents

Factory, officially The San Francisco AI Factory Inc., has spent the past year building a platform that handles complex programming tasks such as porting an application to a new language. Its approach is to let developers type a high-level description, then break that work into sub-steps assigned to a group of AI agents. The platform automatically provisions infrastructure for each agent, prevents agents from interfering with one another's work, and tracks progress in real time.

Several features reflect lessons from running agents on real code. A slash command called Readiness Report checks whether a repository is agent-friendly, because AI agents struggle with code files that are not well-organized or lack technical explanations. Factory also checks for guardrails that prevent agents from introducing vulnerabilities. Once a task starts, each step goes to the large language model that can do it most cost-efficiently, and requests are rerouted if a model has technical issues. Agents run in virtual environments called Droid Computers, which save the libraries and configuration files so developers do not have to install them each session. In January, Factory said it was using LLMs to automatically improve its agents based on anonymized usage signals from thousands of daily sessions.

The new capital points to an expansion phase. Factory disclosed today that hundreds of thousands of developers use the platform, some at Nvidia, Adobe and Palo Alto Networks, and will reportedly grow headcount to 300 by year's end. Whether the company can turn that developer reach into durable enterprise contracts is likely to be the question investors watch, as chief executive Matan Grinberg frames the market as a move from individual coding agents to software factories that underpin an entire software company.