AI Coding Agent Features: Agent Swarm, AI Code Review & PRD Execution
Everything Factory Nexus does: an AI engineering team that decomposes features, writes code in parallel, reviews every PR, and merges to your repo. Here's how each piece works.
Your AI engineering team
Agent Swarm
Describe a feature and Factory decomposes it into parallel tasks. Up to 5 AI agents work simultaneously, each on its own isolated git branch. No merge conflicts, no stepping on toes — just parallel output.
- AI decomposes features into subtasks automatically
- Each agent works in an isolated git worktree
- Up to 5 agents running in parallel per project
- Smart retry with feedback when reviews fail
Every PR reviewed by 3 AI experts
AI Code Review
When an agent finishes, three concurrent AI reviewers check the code for architecture quality, security vulnerabilities, and style issues. Failed reviews trigger automatic retry — the agent gets the feedback and tries again.
- Architecture, security, and style reviews run concurrently
- Verdicts: approve, request changes, or comment
- Auto-retry on rejection with review feedback injected
- GitHub Check Runs and PR comments via GitHub App
From spec to shipped, automatically
PRD Execution
Write a PRD, hit execute, and Factory turns every task into a parallel swarm job. Track progress per-PRD with grouped task cards, status breakdowns, and review verdicts. Your spec becomes working code without copy-pasting tickets into an AI.
- One-click execution of all PRD tasks via agent swarm
- Grouped progress view: per-PRD task counts and status bars
- Each task gets AI code review before merging
- ROI tracking: see time and cost saved per execution
Works where you already work
GitHub Integration
Import any GitHub repo in one click. Label an issue with "factory" and agents start working automatically. Get PRs back on your repo with reviewed, tested code. Factory meets you where you already are.
- One-click GitHub repo import
- Label an issue "factory" to trigger orchestration
- Automatic PR creation with reviewed code
- Issue cards surface GitHub issues as orchestration prompts
Visual CI/CD, built in
Workflows & Pipeline Studio
A full visual pipeline editor for building multi-stage workflows. Define stages with gates (manual approval, scheduled, API callback, conditional), run action steps in Docker containers, and monitor execution in real time.
- Visual YAML pipeline editor with drag-and-drop
- 4 gate types: manual, scheduled, API callback, conditional
- 7 action step types: bash, Python, Node, Go, Rust, curl, custom
- Docker-based test and build runners for 12 languages
Know before you build
Idea Estimator
Paste any project idea and get a detailed AI-generated technical estimate in seconds. Tech stack recommendations, complexity scoring, component breakdown, and a timeline comparison between solo development and Factory agents.
- AI-powered tech stack recommendations
- Complexity score with reasoning (1-10 scale)
- Component breakdown with effort estimates
- Solo dev timeline vs Factory agents comparison
Frequently asked questions
- How does the agent swarm decide what each agent works on?
- Factory decomposes your description into subtasks, assigns each one an owner and a file scope, and dispatches them to separate agents. Tasks that depend on earlier work queue behind it and start as soon as a slot frees up.
- What does the AI code review actually check?
- Architecture quality, security vulnerabilities, and style — three reviewers running at the same time. Each returns approve, request changes, or comment, and rejections are fed back to the agent for a focused retry.
- Can I run this against an existing production repository?
- Yes. Import the repo, and agents work on branches off it. Nothing lands on your default branch without a pull request you merge yourself.
- What happens when a review fails repeatedly?
- The agent retries with the reviewers' findings injected into its prompt. If it still can't pass, the task fails with the reason recorded, and you can retry, re-scope, or take it over manually.
- How many agents can run at once?
- One concurrent agent on the Free plan and up to five on Pro. Extra tasks queue and start automatically as agents free up.
Ready to ship faster?
Describe what you need. Your AI team handles the rest.