What Cursor Agents can automate for Acme.
6 interactive demos showing agent automations saving your engineers time, and harness optimizations improving outputs while reducing token costs.
Build robust automations across every business unit.
Cursor Automations are headless agents that can be triggered by alerts, webhooks, Slack/Teams messages, schedules, and operate like members of your team.
Overnight data job failure
A 2 a.m. failure is no longer a 9 a.m. problem. The auto-Agent diagnoses the issue, patches it, extensively tests the fix, and pushes a PR for human review.
Three tasks, automatically completed by three agents, running on 3 separate cloud instances. All at once.
Every Cursor cloud agent gets its own computer in the cloud, with a terminal, a browser, and a mouse. Send work from Slack or Teams, the web, or your phone. The agents set themselves up, work in parallel, and come back with finished pull requests. Click any desktop to look over an agent’s shoulder.
PriyaEng Director9:14 AM
Before Friday: backfill the orders mart for last week's outage window, add an accepted-values test for promo_kind, and someone please fix the flaky pipeline-health dashboard e2e test 🙏@Cursor
Cloud agents dispatch from wherever work shows up — this one starts in Slack.
The agent proves its work, on camera.
Before a Cursor agent turns in its work, it starts your app, clicks through the feature it just built, and records the session. Your reviewers watch a 30-second video instead of reading 500 lines of code. Watch the proof stack up as the agent checks each flow.
Ready for verification
Backfill the orders mart for last week's outage window
cursor/orders-mart-backfillon Acme/analytics
code complete · 0 evidence · PR blocked until verified
Evidence locker · 0/6
Cursor agents must prove a change works before the PR opens — in a real browser, recorded, every time.
Every code change reviewed before a human ever sees it.
Bugbot reviews more than 2 million code changes a month and holds the highest published fix rate of any AI code reviewer. Watch it scan a change headed for Acme’s codebase, catch three real bugs, and fix them automatically. Then see what that means for the bugs that reach your users.
feat: usage export service + retry queue #1862
feat/usage-export-martsDdevScripted playback of a real review — three genuine bug classes, pinned where they live.
What did all of that just cost Acme? Watch the router.
Most requests don’t need the biggest model. Cursor’s Auto router sends each request to the model that fits the job, and charges one flat rate even when it reaches for a premium model. Replay every request from this page through the router, then drag the sliders to size it for your team.
Usage replay ready
12 requests12 requests captured from this page’s story — the 212-model dbt triage, the fleet’s three builds, the parity checks, the MERGE review. Watch the router Acme would actually use switch each one onto the right model lane.
all-frontier list price for this set: $11.00 · watch what Auto actually bills
Auto picks the model per request and bills a flat Auto rate — even when it upgrades you to a frontier model.
Everything behind this page is yours to own and inspect.
The skills the agent loads per task, the helper agents that work in parallel, the guardrails that control what agents can do, and the connections into your tools are all configuration Acme controls — not settings locked inside Cursor. Toggle the rings, run a task through them, and open any piece to see exactly what it does.
The harness · live config
Flip a layer, then run the task — the run honors whatever is on when it starts.
Skills
4/4 onSOPs the agent loads per task
Subagents
onplanner + 3 specialists, in parallel
Hooks
3/3 ongates on every agent action
“Triage the failed nightly build and post a summary”
this whole panel = 9 pieces of config you own
Every node is a file — click any of them to read it.