Our read
Decide on two points first: where agents must run and how you want to pay. Augment Code documents agents on your own machines or cloud, human approval gates and one pooled usage balance. Cursor suits teams already working in its editor who want cloud agents, Slack-started PRs and Bugbot review on per-tier and per-user plans. The sources leave Cursor's repository indexing and Augment's plan-by-plan admin controls open, so settle those in a pilot on one real ticket queue.
Cursor and Augment Code on the ticket-to-PR path, as each vendor describes it
| Criterion | Augment Code | Cursor |
|---|---|---|
| Execution environment | Augment Code Cloud sandboxes, own machines or both Customer cloud: AWS, GCP named | Cursor Cloud agents on their own computers Sandbox and customer-cloud detail not given |
Augment CodeCosmos agents run in managed cloud sandboxes, on customer machines or in pools of both.[1] CursorCloud agents build, test and demo in parallel for hours or days. Isolation model is not described on the homepage.[2] | ||
| Agent actions | Augment Code Triage, PR author, review, tester Humans approve at gates | Cursor PR from Slack, Bugbot, marketplace automations Cloud agents and Bugbot from Premium |
Augment CodeCosmos experts cover triage, PR authoring, review and testing, with work pausing at quality gates.[1] CursorAgent opens PRs from Slack requests, Bugbot reviews PRs, and the marketplace lists ticket, CI and Slack automations.[2] [3] | ||
| Team controls | Augment Code SOC 2 Type II, ISO 42001; SSO/SCIM on Enterprise Plan table is ambiguous on audit trails | Cursor SOC 2, ISO 27001/42001; SSO on Teams, SCIM on Enterprise Privacy Mode stops training when enabled |
Augment CodeNo training by default. Plan cards place SSO, OIDC, SCIM and CMEK under Enterprise.[4] [5] CursorTeams adds SAML/OIDC SSO and privacy settings; Enterprise adds SCIM, audit logs and access controls.[6] [7] | ||
| Pricing | Augment Code $20 or $100 flat, pooled usage, up to 50 seats 40% fee on model usage; Enterprise on quote | Cursor Tiers $0 to $200; Teams from $30 per user Third-party models billed separately |
Augment CodeStandard 20 and Business 100 dollars a month for the whole team; the same amount of usage is included once per plan, pooled, with top-ups.[5] CursorFree, Plus 8, Premium 30, Super 100, Ultra 200 a month; a free plan exists.[7] | ||
Sources
Where the agent runs matters more than the model list
Augment Code states that Cosmos agents run in managed cloud sandboxes, on your own machines for code that cannot leave the building, or in pools of both. AWS and GCP are named for customer-cloud setups. Triggers include GitHub, GitLab, Slack, Linear, Jira, schedules and typed webhooks.
Cursor's homepage says cloud agents use their own computers to build, test and demo features in parallel, for hours or days. It does not describe the isolation model or say whether those machines can sit in your cloud. Brex describes custom self-hosted Cursor agents on dedicated nodes with access to its cluster, so the pattern exists in at least one deployment. That is a customer's setup, not a statement of what your plan includes. If code cannot leave your network, settle this with Cursor first.
What each agent does between ticket and PR
Cosmos ships experts for each stage: a work dispatcher that scans open tickets and applies a triage rubric, a PR author, a code review fleet and a tester that posts results with screenshots. Work pauses at quality gates, and consequential actions are approved by a person. The IDE side is covered by Visual Studio Code and JetBrains extensions plus a CLI.
In Cursor, a teammate can ask the agent in Slack to take a task and get a PR link back, and Bugbot reviews PRs in GitHub. The marketplace lists automations such as fixing bugs reported in a Slack channel, triaging Linear issues and opening PRs for easy fixes, and fixing CI failures on main. One template auto-approves low-risk PRs; check any such template against your merge policy before enabling it. Cloud agents and Bugbot start at the Premium tier.
Two customer workflows, not a benchmark
Brex wrote that a Linear ticket triggers a PR-generation agent in Cursor, and a PR label starts a second agent that watches CI, deploys to development and fixes failures, capped at three attempts. Agents never merge; a human reviews. Brex reports one engineer migrating 180 microservices in a month, a 32x per-capita gain on one framework upgrade, against about 50 services by three engineers in a quarter earlier. It also reports that some PRs passed tests yet caused runtime bugs where coverage was thin.
MoneyGram's vendor case study is IDE-centred: engineers use Augment in IntelliJ and VS Code with Jira and Confluence through MCP, and report Spring Boot migrations cut from weeks to hours. Pure Storage chose Augment for its 2,000 engineers after Copilot struggled in a monolithic C++ repository, citing the context engine and enterprise security policies. These are different tasks and company-reported results. They show the shape of each workflow, not which tool is better.
Team controls: tier-by-tier on one side, thinner on the other
Cursor's pricing page assigns controls by plan: Teams adds SAML/OIDC SSO, usage analytics and team-wide privacy settings; Enterprise adds SCIM, audit logs, service accounts, spend limits and repository, model and MCP access controls. Its security page lists SOC 2 Type II, ISO 27001 and ISO 42001, and says Privacy Mode, when enabled, means no training on your data.
Augment lists SOC 2 Type II and ISO 42001 and says data is not used for training by default. Its plan cards place SSO, OIDC, SCIM and CMEK under Enterprise, while the comparison table repeats the security block in every column, so confirm which plan carries audit trails and access controls.
Pricing: a pooled balance against tiers
Augment Standard is 20 dollars a month and Business 100: one flat price for the whole team, up to 50 seats, with 20 or 100 dollars of usage included once a month for the plan as a whole, not per seat. Usage covers model inference at provider list price plus a 40% service fee on that inference, and Cosmos compute. The balance is pooled, with top-ups. Cost follows work done, not headcount.
Cursor lists Free, Plus at 8, Premium at 30, Super at 100 and Ultra at 200 a month, plus Teams from 30 per user. Plus is limited to Grok models, and third-party models are billed separately at API list price. Prices can change, so reopen both pages before budgeting.
What to test before choosing
Editorial suggestion, not a vendor claim: run one real queue of tickets through each tool, with the execution location you actually need, and record where a person had to step in and what the month cost. Cursor's secure codebase indexing is listed on its homepage without detail, so test repository context on your largest repo. For a wider field including Warp and Factory, see our four-way comparison.
Method and limits
2026-10-27
This dated snapshot counts distinct companies in public customer evidence. CONFIRMED means a case study, named testimonial or press source; DECLARED means the company itself told usedby; OBSERVED means a vendor listing; SIGNAL means a job post. All four count, but their strength differs. Removed and unlevelled lines are excluded. Examples are a capped selection verified within 90 days. Vendor facts are sourced records checked within 60 days. Coverage is uneven and missing evidence is unknown. These figures measure neither market share nor product performance; capture dates are not adoption dates.
