Factory raises $150M to help enterprises manage fleets of AI coding agents more efficiently

Factory raises $150M to manage enterprise AI coding fleets

Factory, the AI coding agent platform founded by Matan Grinberg, has raised $150 million from Coastal Ventures, Sequoia Capital, Blackstone, Insight Partners, NEA, and additional partners. The company has 70 employees and is opening a London office with part of the proceeds.

The problem Factory is selling against

Large enterprises that pushed developers to adopt agentic AI are now discovering the spend has gotten away from them. Grinberg describes companies burning hundreds of thousands of dollars a month on developers prompting frontier models for trivial tasks — essentially paying top-tier inference rates to say hello to a model. Factory sits in the middle as a model-agnostic routing layer, directing each task to the appropriate model rather than defaulting everything to the most expensive one.

The broader enterprise adoption pattern Grinberg describes has three phases: initial reluctance, a kitchen-sink phase where developers are told to use AI indiscriminately, and a rationalization phase where companies figure out what actually moves the needle. Uber and Meta are, in his view, already in phase three.

“We raised a $150,000,000 from the great folks at Coastal Ventures, Sequoia Capital, Blackstone, Insight, NEA, and some other great partners. Large enterprises that we work with found that on a monthly basis, they spend on the order of hundreds of thousands of dollars on developers saying things like 'hello' to Opus 4.6 fast — which is probably not the best use of those tokens.”

Where the ROI actually is

The projects generating real returns are not greenfield apps. They're the low-leverage work that senior engineers hate: documentation, testing, and legacy code migrations that can consume years of engineering time. Grinberg's framing is that well-paid, highly skilled engineers shouldn't be doing any of it. Factory's "Droid" agent automates those tasks, and the highest-value enterprises are the ones systematically playing whack-a-mole against whatever lowest-leverage work their developers are still doing.

The Tesla factory analogy is where Factory's name comes from — mostly robotic arms, humans designing the system rather than running it.

Model strategy

Factory builds model-agnostic by design, and Grinberg treats that as non-negotiable for serious enterprise customers. A single provider dependency is a reliability risk: if that API endpoint goes down for a regulated industry that has gone fully agent-native, the workflow stops. Different models also perform differently by language and task type, so optionality matters operationally.

On training its own model, Grinberg is direct that it doesn't make sense yet. The incremental alpha in Factory's product comes from the agent layer — Droid — which he says already outperforms agents coming out of the model labs on benchmarks. Fine-tuning a proprietary model would be a lower-return use of engineering hours than improving the agent itself, at least for now.

Deployment and international expansion

Factory's forward-deployed engineers are explicitly not a services business. They work on-site with enterprise engineering teams to identify friction points, then feed those back into the product rather than solving them repeatedly as one-off engagements. The team is positioned as product intelligence, not a consulting arm.

The London office targets European and Asian markets, where data residency and model locality requirements make fully on-premises deployment attractive. Factory supports full on-prem deployment — Grinberg notes it could run on a nuclear submarine with GPUs — which sidesteps the regulatory friction that has slowed other AI developer tools in those markets.