Factory is what you build when you think autocomplete is the wrong level of abstraction. Instead of an AI that sits next to your cursor suggesting one line at a time, Factory gives you a small army of role-specialised agents — called Droids — that own whole tasks from ticket to merged PR. This is a tool for engineering teams that have stopped asking “how do I type faster?” and started asking “how do I delegate to something that isn’t a person?”

It’s also, as of this week, a $1.5 billion company.

New in April 2026: $150M Series C at a $1.5B valuation

On April 16, Factory announced a $150 million Series C led by Khosla Ventures with participation from Sequoia, Insight, Blackstone, NEA, 20VC, and others. That’s a roughly 5x valuation jump from the September 2025 round, and it arrived with a revenue claim the investor class actually cares about: revenue has doubled month-over-month for six consecutive months. Customers named publicly include Nvidia, Adobe, Palo Alto Networks, Morgan Stanley, EY, Adyen, MongoDB, Bayer, and Zapier. Droid also posted a 58.75% score on Terminal-Bench, which as of this writing is the state of the art for an end-to-end coding agent.

The round matters less for the money and more as a signal: investors are betting that enterprise engineering budgets are moving from IDE plugins (where Cursor and GitHub Copilot live) to agent platforms (where Factory, Devin, and Claude Code’s Routines live). If that thesis is right, the competitive map for the next 18 months gets rewritten.

What Droids actually do

A Droid is a role-scoped agent. The Code Droid writes features. The Review Droid reviews PRs. The Test Droid writes tests. The Knowledge Droid indexes your repo, docs, and ticket history so every other Droid shares a common context store. You orchestrate them from VS Code, JetBrains, a terminal CLI, or the web — or you let them run headless against GitHub issues and Linear tickets.

The important design decision is that Droids are model-agnostic. You can point them at Claude Opus 4.7, GPT-5.4, Gemini 3, DeepSeek, or bring your own key. For enterprises with procurement constraints or data residency requirements, that flexibility is the whole pitch. For individuals, it means you can shop for model quality per task without rewriting your workflow.

Where it’s a fit

Factory pays off when three things are true: your team uses a structured ticket system (Linear or Jira, not ad-hoc GitHub issues), you have a real codebase that benefits from persistent context, and you’re willing to rewire your PR review process around agents rather than alongside them. Teams that fit that profile report genuine cycle-time compression — customer quotes in the press coverage put it at roughly 2x feature throughput.

Where it isn’t

If you’re a solopreneur shipping a SaaS MVP, Factory is overkill. Lovable or Bolt will give you a working app faster, and you won’t use the agent orchestration machinery. If you’re an individual developer who just wants autocomplete plus the occasional multi-file edit, Cursor or Windsurf is a cleaner fit. Factory is built for teams, and the value compounds with team size and repo maturity.

Droids also still hallucinate — the press coverage is honest about this. On vague prompts or sparsely-documented repos, output quality drops. Knowledge Droid mitigates this, but only after you’ve invested in the indexing setup.

Pricing

Free tier: 20M tokens, no credit card. Pro starts at $20/month. Overage pricing is $2.70 per million standard tokens, with cached tokens 90% cheaper. Enterprise is custom, with SSO, SAML/SCIM, audit trails, and on-premise deployment available. Bring-your-own-key works on the free plan, which is unusual and useful.

Verdict

Factory is the clearest expression yet of “delegate, don’t type.” If your team is structurally ready to hand work to agents — tickets, code review, testing — it’s one of the two or three serious options alongside Devin and Claude Code. If your team isn’t ready, start smaller. The tool won’t compensate for the workflow discipline it assumes.

Was this helpful?