Five Months, Three Valuations: Factory's Droid Agents Triple to $5B as Enterprise Coding Spend Accelerates

On September 15, Factory — the startup behind the “Droid” autonomous coding agents — announced a $200 million round at a $5 billion valuation, with backing from Blackstone, Khosla Ventures, Sequoia Capital, Insight Partners, Evantic Capital, and Sound Ventures, among others. The round brings the company’s total funding to more than $400 million, and it lands just five months after an April Series C that valued Factory at $1.5 billion. In other words, the company more than tripled its valuation in roughly 150 days.

The Wall Street Journal first reported the financing, and Reuters confirmed the details: the San Francisco-based company raised $150 million at a $1.5 billion valuation in April, meaning this new round represents both a larger check and a dramatically higher price. Factory is the latest agentic-coding startup to see its valuation run ahead of revenue as enterprise buyers ratchet up spend on autonomous-coding subscriptions — but the specifics of what Factory sells explain why investors keep paying up.

What Factory actually builds

Factory was founded in 2023 by Matan Grinberg and Eno Reyes on a thesis that writing code is only a small fraction of what software engineering involves. The real bottlenecks are testing, reviewing, documenting, and deploying code reliably. That thesis became Droid, an agent that operates across the full software development lifecycle: it plans changes, writes code, runs test suites, catches edge cases, writes documentation, and manages CI/CD pipelines — working through a CLI, a web interface, Slack and Teams, Linear and Jira, and mobile.

Two design choices distinguish Factory in an increasingly crowded market.

First, Droid is aggressively model-agnostic. The platform dispatches tasks across foundation models of varying strength — Claude, DeepSeek, and others — routing each unit of work to whichever model offers the right trade-off of capability, cost, and compliance. A --worktree flag even lets developers run parallel Droid jobs against the same repository, each pinned to a different model. As frontier models commoditize, this routing layer insulates Factory’s customers from volatility in the underlying model market: when a newer, cheaper, or more capable model ships, clients plug it in rather than rebuild workflows.

Second, Factory leans into “self-improving” software development — the explicit rationale for this round. The homepage now reads “make your software improve itself.” Beyond one-shot tasks, the company ships Missions, in which Droids pursue goals autonomously over multi-day horizons while an orchestrator breaks projects into pieces, and custom Droids — reusable subagents defined in Markdown, each with its own system prompt, model preference, and tool policy. The pitch is not a smarter autocomplete but a persistent engineering workforce that maintains the infrastructure agents need to operate reliably: clean documentation, robust test coverage, well-kept pipelines. Factory calls this “paving the roads.”

Enterprise traction and benchmarks

The customer list reads like a who’s-who of enterprise technology: Nvidia, Adobe, EY, Palo Alto Networks, Adyen, MongoDB, Bayer, Zapier, Morgan Stanley, RBC, T-Mobile, HP, DoorDash, and Groq have all integrated Droids into development workflows, with hundreds of thousands of developers reported on the platform. Notably, Nvidia is both an investor — it participated in Factory’s $50 million Series B in September 2025 alongside NEA, Sequoia, and J.P. Morgan — and a customer, a dual role that signals real infrastructure-grade usage rather than pilot-program tourism.

On benchmarks, Droid reached number one on Terminal-Bench, the challenging general software development evaluation, with a state-of-the-art 58.75 percent score in September 2025, and independent reviews in early 2026 put it at 63.1 percent — ahead of OpenAI’s Codex CLI at the time. Factory agents have occupied multiple positions in Terminal-Bench’s top five even as the leaderboard churns with each new frontier model release.

The growth numbers explain the valuation arithmetic. At the April Series C, Factory reported revenue doubling every month for six straight months. Even if that pace has cooled, the trajectory from $1.5 billion to $5 billion in five months reflects investors underwriting continued hyper-growth in enterprise agent subscriptions.

The agentic coding funding wave

Factory’s round does not exist in isolation. It is the third eye-popping agentic-coding financing in barely a week:

• Cognition, maker of the Devin agent, raised $2 billion at a $48 billion valuation in early September, with annualized revenue above $900 million and projected to reach $4–5 billion by end of 2026.
• Cursor (Anysphere) has been reported seeking around $2 billion at a $50 billion valuation.
• Factory now closes $200 million at $5 billion.

Three companies absorbing the bulk of AI venture capital, as one widely-shared summary put it. The pattern across all three: valuations are running ahead of revenue, and investors are explicitly betting that autonomous coding is far from a winner-take-all market.

What links them strategically is a shared thesis — the harness is the product, and the model is a line item. As foundation models improve in bursts and prices fall, the durable value accrues to the orchestration layer: the routing, the guardrails, the enterprise integrations, the audit trails. Factory’s model-agnostic Droids are perhaps the purest expression of that bet. If next quarter’s frontier model halves the cost of a token, Factory’s margin story improves without the company rewriting a line of its core product.

Risks worth taking seriously

The bull case is clear, but the risks are real. First, the labs themselves are competing: Anthropic’s Claude Code is the most heavily used agentic coding tool on the market, and OpenAI’s Codex line ships with distribution advantages no startup can match. A routing layer is only defensible while the models beneath it stay genuinely substitutable.

Second, demand is showing signs of strain at the margin — three major labs rationed coding capacity within a single week in September, and pricing models are in flux. Subscription economics that work at 100,000 developers may need rethinking at a million.

Third, the class action facing Anthropic over how subscription limits are marketed at the point of sale is a reminder that agentic-coding pricing will attract regulatory and legal scrutiny as spend concentrates.

What to watch

The $5 billion question is whether Factory’s enterprise contracts convert into the kind of durable, expanding revenue that Cognition has demonstrated. Watch three signals: whether Droid holds top positions on Terminal-Bench as labs ship purpose-built agentic models; whether the Missions and custom-Droid surface area deepens switching costs beyond what a model provider can replicate natively; and whether the April “revenue doubling every month” trajectory shows up in a credible ARR disclosure ahead of an inevitable next round.

One thing is certain: five months is now enough time to triple a valuation in this market. The agentic coding layer is where the money is moving, and Factory just claimed its seat at the table.