Droid (Coding Agent)

by factory-dot-ai (Droid)

Autonomous software development agent that handles tasks from specification to deployment

See Factory Droid docs and demos

Summary

Droid is a full-stack coding agent that goes beyond code completion to perform end-to-end software development tasks: ticket triage, feature implementation, debugging, CI/CD integration, and incident triage. It emphasizes planning, multi-step execution, and integration with developer tooling.

Features

  • Ticket-to-code workflows: convert tickets/PR comments into implementable plans and patches
  • Integration with GitHub/GitLab, Jira/Linear, Slack, PagerDuty for context and actions
  • Debugging and incident triage with context-aware fixes and rollbacks
  • Multi-resolution codebase understanding (HyperCode) and retrieval (ByteRank)
  • Sandbox execution and fine-grained guardrails for safety

Superpowers

Enables teams to offload repetitive engineering tasks, accelerate onboarding, and maintain persistent agent memory across long-running projects.

Status and plans (secondary sources, 2026-09-30)

  • Factory raised a 1.5B valuation on 2026-04-16, led by Khosla Ventures (several outlets; no Factory primary page fetched).
  • Droids run across desktop, CLI and SDK; Factory sits at the top of the Terminal-Bench leaderboard per press coverage (leaderboard not checked).
  • Plans reported: Pro, Plus, Max, Business (up to 150 seats) and Enterprise custom.
  • Factory was one of the contributors to the AGENTS.md format (see agents-md).

Droid CLI (terminal surface)

Droid CLI is the terminal entry point of the same agent (natural-language task, plan, code and test across a repository, open a pull request). The agent is also offered in VS Code, JetBrains, a browser and Slack/Teams. The points below come from the earlier droid-cli note (Factory product pages and third-party guides, not re-verified on 2026-10-02):

  • Model flexibility (BYOK): Claude, GPT, Gemini or other providers per task; bring-your-own-key extends to Ollama, OpenRouter, Fireworks and DeepInfra.
  • Spec Mode: a reviewable spec before the agent edits; Missions for longer multi-step autonomous jobs; custom droids, skills and plugins.
  • Adjustable autonomy: from fully supervised to autonomous execution.
  • droid exec: non-interactive mode for CI/automation; native git worktree support for parallel tasks.
  • Droid Control (stable in 2026 per those sources): operates browsers and desktop apps for QA reports and demo videos.
  • Automatic AI code review, security scanning and PR descriptions.
  • Terminal-Bench: ranked #12 with 77.3% (+/-2.2), version 2.0, tested 2026-02-24 (earlier note; leaderboard not re-checked, and the press claim above that Factory tops the leaderboard conflicts with it).
  • Freemium: a free tier and paid plans (see the plan list above).
  • Competes with Claude Code and OpenAI Codex; differentiators are multi-model support and one agent across CLI, IDE, browser and chat.

Known limitations & notes

  • Enterprise integration and guardrails required for safe production usage
  • Not a replacement for human engineers — intended to augment and accelerate workflows

Sources