Devin is Cognition's autonomous AI software engineer: you assign it work through Slack, Linear, Jira, or the web app, and it plans, writes, runs, and tests code in its own cloud environment before opening a pull request. Pricing as of August 2026 is Free, Pro at $20/mo, Max at $200/mo, and Teams at an $80/mo minimum with $40 full seats and free flex seats. Devin replaced its old ACU billing with daily and weekly token quotas in March 2026, with overage billed at API list prices.
- Devin is delegation-shaped. You give it a ticket, not a keystroke.
- Pricing: Free, Pro $20, Max $200, Teams from $80/mo with $40 full seats.
- ACUs are gone. Since March 2026 it is a daily and weekly token quota with metered overage.
- Devin Desktop is Windsurf, renamed on 2 June 2026. Same editor, new name.
- It wins on parallel, well-specified, low-context work. It loses on anything you need to steer.
What Devin is
Cognition describes Devin in its own documentation as an autonomous AI software engineer that can write, run, and test code, built to help engineering teams crush their backlogs. That framing is the product, not marketing gloss: Devin is aimed at the queue of work nobody gets to, rather than at the file you have open.
Mechanically it is a cloud agent with its own environment. You hand it a task, it plans, works, runs tests, and comes back with a pull request. The interesting design decision is where you hand it that task from.
Where work reaches Devin.
| Surface | What it is for |
|---|---|
| Slack and Microsoft Teams | Tag it in a thread and it picks the work up from there |
| Linear and Jira | Assign a ticket the way you would assign a person |
| GitHub, GitLab, Bitbucket | It opens PRs, and Devin Review reviews yours |
| Devin CLI | Terminal surface, if you want one |
| Devin Desktop | The IDE, formerly Windsurf |
| Web app | Sessions, playbooks, automations, scheduled runs |
The rest of the surface
- DeepWiki indexes a repository into a browsable, queryable wiki, which is how Devin builds context on a codebase before it starts.
- Devin Review reviews pull requests. Public GitHub PRs are free at devinreview.com, or by swapping github.com for devinreview.com in any PR URL, with no account required.
- Playbooks capture a repeatable procedure so the same class of task runs the same way twice.
- Automations and scheduled sessions run work on a trigger or a clock rather than on a request.
- Agent Command Center and Spaces, in Devin Desktop, are for running several agents at once and letting them share context.
Pricing
Devin has four self-serve plans plus Enterprise. Prices were read from the Devin pricing page and the self-serve billing documentation on 7 August 2026.
Devin self-serve plans, August 2026.
| Plan | Price | Members | For |
|---|---|---|---|
| Free | $0 | 1 | Trying it. Light quota, limited model availability |
| Pro | $20/month | 1 | One developer using Devin regularly |
| Max | $200/month | 1 | Consistently exceeding the Pro quota |
| Teams | $80/month minimum | Unlimited | Shared work, admin dashboard, analytics |
| Enterprise | Contact sales | Unlimited | SAML/OIDC SSO, VPC deployment, dedicated support |
The Teams floor, which is the confusing bit
Teams costs at least $80 a month whatever you do with it. Full seats are $40 each and include a Pro-equivalent quota plus Devin Desktop access. Flex seats are free, unlimited, and draw entirely from the team's shared on-demand credit pool instead of carrying quota of their own. Anything short of the $80 floor is charged as prepaid credits the whole team can spend.
How the $80 minimum resolves, from Cognition's billing docs.
| Full seats | On-demand credits included | Monthly total |
|---|---|---|
| 0 | $80 | $80 |
| 1 | $40 | $80 |
| 2 | $0 | $80 |
| 3 | $0 | $120 |
| N, where N is 2 or more | $0 | N x $40 |
The quota model that replaced ACUs
If you read about Devin before 2026, you read about Agent Compute Units. That billing unit is gone. In March 2026 Devin replaced the credit-based system with a quota-based usage system, and the documentation now describes plan allowances rather than compute units.
| Old ACU model | Current quota model | |
|---|---|---|
| Unit | Agent Compute Units | Tokens, weighted by model |
| Shape | Buy credits, spend them | Daily and weekly allowance that refreshes |
| Running out | Buy more ACUs | Buy extra usage, or wait for the reset |
| Overage price | Per ACU | API list prices for that model |
| Free models | n/a | Do not count against quota at all |
- Pro and Teams full seats get a daily and a weekly allowance. The daily allowance is deliberately more than one seventh of the weekly one, so weekend work does not cost you the week.
- Max has a weekly allowance only, with no daily cap, which is the actual difference between $20 and $200 rather than a simple multiplier.
- Overage is billed at API list prices for whichever model you used, and priority or speed configurations cost more.
- On-demand credits roll over month to month and purchased credits never expire. On Teams they are shared across every member with no per-person balance.
- Cognition's own advice for stretching quota: be precise, drop unnecessary context, stay on one frontier model so caching works, and use the in-house SWE models that do not count against quota. SWE-1.7 is listed as free through 8 August 2026.
Devin Desktop is Windsurf
On 2 June 2026 Cognition renamed the Windsurf editor to Devin Desktop, shipped as a standard over-the-air update. Its own FAQ is unambiguous: Devin Desktop is the new name for Windsurf, and plan, pricing, extensions, and settings all carry over automatically. windsurf.com now redirects to devin.ai/desktop.
What changed with the name was the default interface. The Agent Command Center became the primary surface, and Spaces let several agents collaborate and share context, which is Cognition folding the fleet-of-agents idea into the editor rather than leaving it in the cloud product.
Where it fits against a terminal agent
Devin and a CLI agent like Claude Code or Codex are not competing implementations of the same idea. They are different bargains about supervision.
| Devin | Terminal agent | |
|---|---|---|
| Where it runs | Cloud environment it owns | Your machine, your repo |
| How you start work | Assign a ticket or tag it in Slack | Type a prompt |
| Supervision | Asynchronous. You read the PR | Continuous. You steer every turn |
| Context it starts with | Whatever it indexed, via DeepWiki | Whatever you point it at |
| Best at | Well-specified, independent, parallel tickets | Ambiguous work where taste matters |
| Worst at | Anything needing a correction mid-flight | Anything you want done while you sleep |
| Entry price | $20/mo | $0 to $20/mo |
What it is genuinely bad at
- Under-specified work. An autonomous agent given a vague ticket produces a confident, wrong PR, and you pay for the whole run before you find out.
- Anything with a lot of tacit context. DeepWiki indexes what is in the repository, not what your team knows.
- Work you want to correct halfway. The asynchronous loop is the point, and it is also the limitation.
- Review capacity. Several agents in parallel move your bottleneck to reading diffs, which no plan tier fixes.
A short history, because it explains the product
- March 2024. Cognition introduces Devin as the first AI software engineer. The demo sets the expectations, high and low, that the product has been arguing with ever since.
- Before 2026. Devin sells self-serve plans billed in Agent Compute Units, the unit every older comparison article still quotes.
- 14 July 2025. Cognition acquires Windsurf, days after Google hired Windsurf's chief executive and co-founder in a separate deal. Cognition gets the editor, the plugins, and the enterprise business.
- March 2026. ACUs are retired in favour of daily and weekly quotas.
- 2 June 2026. Windsurf is renamed Devin Desktop. One brand, three surfaces: cloud, CLI, and IDE.
That sequence is why the product feels broader than "an autonomous agent". Cognition bought an editor and a plugin business, and has spent a year making one brand out of a cloud agent and a desktop IDE.
Questions people ask
Cognition's autonomous AI software engineer. You assign it work from Slack, Linear, Jira, or the web app, and it plans, writes, runs, and tests code in its own cloud environment before opening a pull request.
As of August 2026: Free, Pro at $20 a month, Max at $200 a month, and Teams at an $80 monthly minimum with $40 full seats and free flex seats. Enterprise is custom.
They were replaced in March 2026 by a quota-based system with daily and weekly token allowances. On-demand credits carry the same dollar value as the old ACUs, and legacy Core plan users were moved to the Free plan.
Yes. Cognition renamed Windsurf to Devin Desktop on 2 June 2026 as an over-the-air update, and says plan, pricing, extensions, and settings carry over automatically. windsurf.com now redirects to devin.ai/desktop.
It depends on whether your backlog is full of well-specified independent tickets. Devin is strong at parallel work you can describe precisely and weak at work you need to steer mid-flight. Try it on the Free plan against real tickets before buying seats.
Pro has a daily and a weekly allowance. Max, at $200, has a weekly allowance only with no daily cap, so it suits people whose usage is lumpy rather than evenly spread.
Cognition's repository indexer. It turns a codebase into a browsable, queryable wiki, which is how Devin builds context before it starts working on a task.
No, it moves the work. Delegating well-specified tickets to an agent shifts your time from writing to specifying and reviewing, and review capacity becomes the constraint long before generation does.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Devin pricing plan prices and seat structure
- Devin docs: self-serve plans the $80 Teams floor, seat types, ACU migration
- Devin docs: quota-based usage quota model, overage at API list prices, grandfathered prices
- Cognition: introducing Devin Desktop the 2 June 2026 Windsurf rename