Devin
AI software engineer that runs a fleet of coding agents
1/3Devin is Cognition's AI software engineer, used to plan and carry out engineering work end to end rather than to autocomplete lines: code migrations and refactors across many repos in parallel, pull request review with visual QA in a real browser, issue triage from Slack and Datadog, CI failure fixes, and scheduled chores such as daily QA and release notes. Devin Desktop, the macOS app that grew out of the Windsurf editor, is where local and cloud agents are managed and reviewed side by side, and DeepWiki generates documentation and system diagrams for codebases nobody on the team wrote. It runs Cognition's own SWE model alongside frontier models from Anthropic, OpenAI and Google.
- Cloud agents plan and execute tasks and open pull requests
- Devin Desktop manages local and cloud agents from one surface
- Devin Review does pull request review and visual QA
- DeepWiki generates docs and diagrams for legacy codebases
- Fleets of agents migrate repositories in parallel
- Scheduled chores such as daily QA and release notes
- Issue triage from Slack and Datadog, plus CI failure fixes
- SWE 1.7 included, plus Claude, OpenAI and Gemini models
What people are saying about Devin
44% positive“Devin has a few nice features like a nice model picker UI that shows you model price in a clear way. The context window widget shows you a countdown to when the cache expires and warns you when it pas expiration.”
“We've employed devin's slack bot as internal customer support. It works really well for this! Previously operations, people would tie up a couple hours per question of engineer time.”
“We tried Devin in the prehistoric times of Claude Sonnet 3.5-3.7 when agentic coding went off the rails after like 30 minutes.”
“Devin treats you like a customer, and not like a lab rat, where they destroy your entire workflow on a whim because somebody at Anthropic had the bright idea that the desktop app should no longer permit resuming subagents which already have the right context cached.”
“Following, because it blows my mind that devin still exists.”
“The Devin native integrations give you a good idea of which platforms are still core/critical in an AI-first development org.”
“in the last two months devin has hit pmf with ai-native startups. tons of new signups from my favorite teams”
“I am going to conduct tests of Devin AI ... Let's see what kind limits they have to us”
“I see a small hit in performance, it's less stable than Q8 about the tokens per second, but honestly, it's better to have more accuracy and wait a little longer than a bad performing model and I say this while having a devin cli subscription that gives me a "free and unlimited" GLM 5.2 that really sucks as I explained in my original comment.”
“This type of AI-to-AI review should be done agent-to-agent, not through Github PRs with tools like Devin.In a manual review, I then expect all “machinery” to already be properly reviewed, and can focus on design / architecture.”
“- token cost is way higher than official ant/oai”
“\- Sequential for Deep Reasoning: Single-threaded stateful chains with memory compression (the Devin / Cognition architecture).”
“* Their answer: one single-threaded agent plus context compression. This is how Devin works”
“use it the same way devin fusion works (strong + weak model combination to get same results as strong model for cheaper).”
“AI coding agent PR merge rates: Claude 84%, Codex 74%, Devin 43%, humans 85%”
“essential $200/mo subscriptions: - Codex - Claude - Cursor - Devin - Droid it now costs $1,000/mo to be an entry-level Agentic Engineer”
“Doesn't the 1% include e.g. Devin, Cursor, OpenRouter? Guessing the underlying spend is more distributed”
“Devin for QA is so slept on”
“With the current Anthropic situation and OpenAI pulling out of Cursor, I started using Devin Desktop app from @cognition . Lets me have occasional access to Claude without locking me entirely with Anthropic. So my current coding arsenal is Codex, Devin and Grok Build.”
“Devin has autonomous sessions.”
“@wongmjane @cognition @devindesktop Can Devin build Linear before Linear builds Devin?”
“I had Claude & Devin collaborating. CC was adversarially reviewing a Devin PR, saw issues, cut its own branch with fixes, got Cursor to thumb, then merged it in. 😂”
“Devin Automations is the most obvious one: solid UX, but can get very expensive to run.There are also a bunch of newer startups in this space.”
“Great write up of some recent rendering performance improvements the team has made to @DevinAI’s UI.”
“Being able to hand a task from one specialized agent to another without manually moving the context with it. That sounds small until you’ve tried running multiple agents on the same piece of work.”