What it is
Devin began in 2024 as Cognition's autonomous cloud software engineer: you give it a task in Slack or the web app and it works in its own sandboxed virtual machine, then opens a pull request. In July 2025 Cognition bought the Windsurf editor (originally Codeium), and on 2 June 2026 it renamed Windsurf to Devin Desktop. windsurf.com now redirects to devin.ai, and the product line is Devin Cloud, Devin CLI, Devin Review and Devin Desktop, all on one set of plans.
Devin Desktop is still a full VS Code-style IDE with Tab completions, but it opens on an Agent Command Center, a board that tracks every local and cloud agent session by status. Its local agent, Devin Local, replaces the old Cascade agent and shares a Rust-based harness with Devin CLI. Through the Agent Client Protocol (ACP) it can also run outside agents such as Codex, Claude Agent and OpenCode on Pro, Max and Teams plans.
October 2026 added Devin Oncall, which investigates alerts and bug reports, cross-session memory, and an Efficiency Score that lets admins hand bigger AI budgets to engineers who use them well. The Windsurf JetBrains plugin keeps its old name.
Our review
Devin sells a local IDE and a hand-it-a-ticket cloud agent on the same $20 plan, and that is the whole case for it. The Agent Command Center shows Devin's own sessions next to Codex or Claude Agent running through ACP, so a team can keep one window open whichever model wins this month. The product is also widening beyond coding: Oncall for production alerts, a Pylon hook for support teams and a Microsoft Teams integration all shipped between September and October 2026.
The execution has been messy. DevAI Reviews, looking at Devin Desktop in August 2026, found the new Devin Local agent could ship switched off while new tabs fell back to Cascade, so many users were still on the old agent without knowing. Cognition's claim that Devin Local is up to 30% more token-efficient has not been independently checked, and its 28 September post cut the cost of Fusion and Normal sessions by 30 to 40%, which tells you how loosely the earlier pricing was set.
Budgeting is the real problem. The pricing page says quotas reset daily and weekly and vary by model, task size and reasoning effort, but gives no numbers for any tier, including the $200 Max plan. Overage is billed at API rates that the page also leaves out. Cognition's own advice is to switch to smaller models such as Haiku or Kimi K2.5 for routine work, which is sensible but shifts the cost-control job onto each engineer.
Security is better documented than usage. Cognition holds a SOC 2 Type II report from September 2024, its docs say it does not train on customer code by default, and each cloud session runs in a fresh virtual machine that is torn down afterwards. The catch is that SAML SSO and dedicated or VPC deployment sit on Enterprise, so a Teams customer gets an autonomous agent with repository credentials but signs in with ordinary accounts.
Teams pricing works out to $40 a seat plus an $80 base, so a ten-person team pays $480 a month. Whether that is a bargain depends on a quota nobody outside Cognition can see, and the new Efficiency Score, which rations budget by how well each engineer uses it, suggests Cognition expects heavy users to run out.
Where it earns its place
A ten-engineer SaaS team on GitHub and Linear
Clear the backlog of small bugs without pulling people off feature work
Engineers tag Devin on Linear tickets; it works in a cloud VM and opens a pull request for review. Teams costs $80 plus $40 a seat, $480 a month, but quotas are unpublished, so heavy users may hit the weekly cap and spill into API-rate overage.
An SRE lead at a company with a PagerDuty-style on-call rota
Get a first diagnosis on alerts before a human is paged
Devin Oncall, generally available since 15 October 2026, investigates alerts and bug reports and hands the engineer its findings. It needs repository and log access, so check whether your plan includes SSO and audit controls (Enterprise) before connecting production systems.
An engineering director at a 40-person team
Put a budget on agent use before rolling it out
Teams costs $1,680 a month at list price, but Devin publishes neither per-seat quota sizes nor overage rates. The October 2026 Efficiency Score and approval tools let admins give bigger budgets to engineers who use them well, which helps control spend but not forecast it.
Before you commit
- Setup
- Download Devin Desktop for Mac, Windows or Linux, or connect GitHub, GitLab or Bitbucket and Slack for Devin Cloud; Windsurf installs updated in place, with plan, pricing and extensions unchanged.
- Integrations
- Slack, Microsoft Teams, Linear, Jira, GitHub, GitLab, Bitbucket, custom git hosts, Pylon and a Devin API; ACP for outside agents on Pro, Max and Teams.
- Data and privacy
- SOC 2 Type II since September 2024; docs say customer code is not used for training by default and data is kept only for the customer relationship, though feedback data may be retained. SAML/OIDC SSO and dedicated or VPC deployment are Enterprise only.
- Cost at scale
- A 25-seat Teams account pays $80 plus $1,000, so $1,080 a month, before overage; overage rates and per-seat quota sizes are not published.
- Limits
- Daily and weekly quotas with no published numbers; cost per message varies by model and task size, and the free plan has no cloud agents.
What's good
- One plan covers an IDE, a CLI and an autonomous cloud agent
- Agent Command Center tracks local and cloud sessions together
- Runs Codex, Claude Agent and OpenCode through ACP on paid plans
- Cloud sessions run in throwaway VMs; no training on code by default
- Unlimited Tab completions on the free plan
What's not
- No published quota numbers on any plan, including $200 Max
- Overage billed at API rates the pricing page does not list
- SSO and VPC deployment are Enterprise only
- Cascade-to-Devin Local switch caused silent fallbacks
- Renamed, re-priced and re-platformed repeatedly since 2025
Best for
Teams that want an AI IDE and an autonomous cloud coding agent from a single vendor.
Alternatives
Devin (incl. Devin Desktop, formerly Windsurf): common questions
- Is Devin (incl. Devin Desktop, formerly Windsurf) worth it?
- A capable editor plus cloud agent under one subscription, but a year of renames, agent swaps and unpublished quotas has left it harder to plan around than Cursor or Claude Code. We rate it 4.1 out of 5.
- How much does Devin (incl. Devin Desktop, formerly Windsurf) cost?
- Free (light agent quota, unlimited Tab completions, no cloud agents); Pro $20/month; Max $200/month; Teams $80/month base plus $40 per full seat, up to 200 users; Enterprise custom. Quotas reset daily and weekly with no published numbers; extra usage billed at API rates.
- What are the downsides of Devin (incl. Devin Desktop, formerly Windsurf)?
- No published quota numbers on any plan, including $200 Max; overage billed at API rates the pricing page does not list; sSO and VPC deployment are Enterprise only; cascade-to-Devin Local switch caused silent fallbacks; renamed, re-priced and re-platformed repeatedly since 2025.
- Who is Devin (incl. Devin Desktop, formerly Windsurf) best for?
- Teams that want an AI IDE and an autonomous cloud coding agent from a single vendor.
- How much does Devin (incl. Devin Desktop, formerly Windsurf) cost for a team or in production?
- A 25-seat Teams account pays $80 plus $1,000, so $1,080 a month, before overage; overage rates and per-seat quota sizes are not published.
- Is Devin (incl. Devin Desktop, formerly Windsurf) safe for company data?
- SOC 2 Type II since September 2024; docs say customer code is not used for training by default and data is kept only for the customer relationship, though feedback data may be retained. SAML/OIDC SSO and dedicated or VPC deployment are Enterprise only.
- What are the best Devin (incl. Devin Desktop, formerly Windsurf) alternatives?
- The closest alternatives we have reviewed are Cursor, Factory (Droids), OpenAI Codex and Claude Code.
Answers checked Oct 23, 2026.
Make Devin (incl. Devin Desktop, formerly Windsurf)? Show this rating on your site
