Cognition
Also known as: Devin
Maker of Devin, an autonomous AI software engineer. Devin Desktop (June 2026) adds an Agent Command Center, Spaces for multi-agent context sharing, and ACP support, shifting from coding assistant toward multi-agent management.
Cognition is the company behind Devin, one of the first AI systems marketed as an autonomous software engineer rather than a coding assistant. Founded in 2023, Cognition built Devin to take on engineering work end to end: given a task from a ticket or a chat message, Devin spins up its own cloud sandbox with a full development setup, a shell, a browser, and an editor, then plans the work, writes the code, runs and tests it, and opens a pull request for review, all while the developer does something else.
What distinguishes Devin is how much it is built to operate without a human in the loop. It tests its own work using computer use, reviews and fixes its own code, can coordinate teams of sub-agents working in parallel, and can schedule its own jobs. Cognition is candid that Devin is strongest on well-scoped tasks, the kind of clearly specified work that would take a junior engineer a few hours, and treats it as a teammate you delegate to rather than a tool you drive keystroke by keystroke.
In 2025 Cognition acquired Windsurf, the agentic IDE that began as Codeium, giving it a second surface alongside the cloud agent. That product has since been rebranded Devin Desktop. It pairs a full editor, with autocomplete and debugging, with an Agent Command Center that puts local and cloud agents in a single Kanban view of what is running, waiting for review, and done, and Spaces that let multiple agents share context and Git worktrees. It also ships Cognition's own fast coding models, the SWE series, and features like Codemaps for visualizing code structure and Fast Context for retrieving the exact files an agent needs.
Together, the two surfaces reflect Cognition's bet on a single workflow: think and prototype locally, delegate longer implementation, testing, and deployment to a cloud agent, then review the resulting pull request in one place. Devin is used by enterprise engineering teams to parallelize routine development work, and Cognition continues to position it as a step from assisted coding toward autonomous software engineering.
Vendor details
Canonical URL
https://cognition.ai
Category
Coding agent
Company status
independent
Use cases & customers
Target customers
Deployment options
In practice
You have a backlog of well-scoped tickets and no one to work them. You hand one to Devin from a Slack message or Linear issue, and it spins up a sandbox, writes and tests the code, and returns a draft pull request to review.
You want to keep coding locally while longer work happens in the background. Devin Desktop's Agent Command Center shows your local and cloud agents in one Kanban view, so you track what's running, waiting for review, and done.
A migration needs many similar changes done in parallel rather than one engineer grinding through them. Devin can run teams of sub-agents at once, each implementing and testing its piece, then surface the results for review.
Capability coverage
10.5 / 14 capabilities · 75%
| Integrations & Tool CallingAgent Features research report + JSON Feature Rubric | Full |
|---|---|
| Workflow OrchestrationAgent Features research report + JSON Feature Rubric | Full |
| Knowledge Grounding & RAGAgent Features research report + JSON Feature Rubric | Full |
| Human Oversight & GuardrailsAgent Features research report + JSON Feature Rubric | Full |
| Security, Identity & GovernanceAgent Features research report + JSON Feature Rubric | Full |
| Observability & AuditabilityAgent Features research report + JSON Feature Rubric | Partial |
| Memory & State PersistenceCognition blog — Devin Desktop + Spaces launch | Full |
| Deployment & Data ResidencyAgent Features research report + JSON Feature Rubric | Partial |
| Prebuilt Agents, Templates & PacksAgent Features research report + JSON Feature Rubric | Unable to verify |
| Triggers & Channel CoverageAgent Features research report + JSON Feature Rubric | Full |
| Model Flexibility & RoutingAgent Features research report + JSON Feature Rubric | Unable to verify |
| APIs, SDKs & MCP ExtensibilityAgent Features research report + JSON Feature Rubric | Full |
| Testing, Debugging & OptimizationAgent Features research report + JSON Feature Rubric | Partial |
| Browser & Computer UseAgent Features research report + JSON Feature Rubric | Full |
Recent platform changes
Cognition acquired The Interaction Company of California, developers of the conversational AI assistant Poke, for a sum in the low nine figures. The company intends to integrate Poke's proactive, personality-driven interaction architecture directly into the Devin coding agent.
View sourceCognition launched Devin Security Swarm, a multi-agent system on an Agentic MapReduce architecture that scans whole codebases to find, validate, and fix vulnerabilities, reproducing findings in isolated sandboxes and opening patch PRs.
View sourceDevin Playbooks now support a structured output schema so Devin can return defined JSON results.
View sourcePricing
Free · Pro $20/mo · Max $200/mo · Teams $80/mo + $40/seat · Enterprise custom
quota + usage beyond quota
Included quota
Pro, Max, Teams, Enterprise announced; exact included quotas not public in captured excerpt
What is public
Cognition announced a new self-serve lineup for Devin: Free, Pro ($20/month), Max ($200/month), Teams (usage-based with a minimum spend of $80/month), and Enterprise (custom). [9]
Billing mechanics
The post is unusually clear that included plan usage now counts against quota and that any usage beyond quota is billed in dollars rather than ACUs for self-serve users. It also says Ask Devin, Devin Review, and premium DeepWiki modes are moving to usage-based billing. Devin Review gets a 2-week free trial, after which a typical review run is expected to cost about $2–$3; DeepWiki for open-source repos remains free. [9]
Cost watchouts
Usage beyond quota, Devin Review runs, Ask Devin/DeepWiki usage
Variable cost rationale
Usage beyond quota is billed separately, Teams has a minimum spend, and review/knowledge features add run-based cost.
Overage / add-ons
Teams min spend $80/mo; overage billed in dollars; Devin Review typically ~$2–$3/run
Sales call required
No — self-serve available
Free / trial
Free tier: light agent quota, limited model availability, unlimited inline edits and Tab completions.
Lowest paid plan
$20/mo
Key ambiguities
Included quotas not public in captured post
Cancellation / refund
n/p in captured pricing materials
Support SLA / resale
Enterprise available; no public SLA found
Missing data
The captured official material did not expose the included quotas for Pro and Max, a public refund policy, or public contract terms for Teams beyond the $80 minimum. For a website comparison, Devin should be marked public partial. [9]
Related vendors
- 10Web — Agentic website platform whose team of ten specialized AI agents…
- Ageiro — Early stage enterprise platform that turns business goals into…
- Aider — Open-source, model-agnostic terminal coding agent that works…
- Anthropic Claude Code — Anthropic's terminal and IDE-native coding agent with sub-agent…
- AppFactor — Agentic orchestration platform that autonomously maintains and…
- Augment Code — AI coding assistant evolving into an AI-native engineering platform