Overview
Devin is an autonomous software engineer AI agent by Cognition AI. It executes end-to-end software tasks — opening PRs, running tests, and modernizing legacy code — the way a human engineer would, and the same agent is reachable from Web, Desktop, and CLI.
Key Features
1. Devin Fusion (multi-model harness)
Rather than pointing one expensive frontier model at every step, Fusion runs a frontier "main agent" alongside a cheap "sidekick" model, hands mechanical work to the sidekick, and switches models mid-task. It runs behind the scenes on the plan you already pay for, and Cognition reports up to 60% lower cost on its own benchmark.
2. SWE-1.7 (in-house coding model)
An in-house model trained from a Kimi K2.7 base with reinforcement learning for coding. Served via Cerebras at roughly 1,000 tokens/sec, it scores 81.5% on Terminal-Bench 2.1 and 77.8% on SWE-Bench Multilingual, and is available across Devin Web, Desktop, and CLI.
3. Interactive Planning and Parallel Sessions
Refine the plan collaboratively before work starts, and spin up multiple Devins to tackle different tasks concurrently.
4. Devin Search / Devin Wiki
Search and summarize across internal repositories to build an understanding of the codebase.
5. Devin Desktop
A code editor plus AI agent coordination environment. Agent Client Protocol (ACP) support lets Codex, Claude Agent, and other third-party agents run from the same interface.
6. Devin Security Swarm
Multiple Devin agents analyze code in parallel, detect vulnerabilities, validate each finding in an isolated sandbox, and open the patch PR.
7. Legacy Code Modernization
Ingests massive legacy codebases (COBOL, Fortran, Objective-C) and proposes or implements refactors into modern languages.
8. Visual Input Support
Consumes Figma designs, UI mockup images, and screen recordings to understand visual bugs and UI requirements.
Plans & Pricing
| Plan | Monthly | Perks |
|---|---|---|
| Free | $0 | Trial usage |
| Pro | $20 | Individual use, SWE-1.7 included |
| Max | $200 | Heavy usage allowance |
| Teams | $80 + $40/seat | Team features |
| Enterprise | Custom | VPC deployment, SAML SSO |
Self-serve pricing is now four tiers — Free, Pro at $20, Max at $200, and Teams at an $80 base plus $40 per seat — with overage billed in ACUs. An ACU (Agent Compute Unit) is a normalized measure spanning VM run time, inference tokens, and networking — roughly 15 minutes of Devin actively working.
Typical Use Cases
- Automated bug fixes and PR creation
- Legacy code modernization
- Test generation and coverage improvement
- Vulnerability detection with patch PRs
- Converting specs and mockups into working code
Strengths
- Free tier to try, individual use from $20
- Parallel agents for multitasking
- Devin Fusion holds quality while cutting cost
- Handles large legacy codebases
- Accepts Figma and video as input
Weaknesses
- ACU consumption is hard to forecast
- Complex tasks still require human review
Official Links
- Website: https://devin.ai
- Pricing: https://devin.ai/pricing
使いこなし Tips
SNS・コミュニティで共有されている実践的な使い方
- In Devin, add an agents.md file at your repo root — it's read as a project manual (background, rules, conventions) that speeds up initial setup (Source: https://x.com/JulianGoldieSEO/article/2079822793454358898)