Build an Agent for Your Industry
Agents don’t just answer. They act, writing and running real code against real systems instead of calling a fixed menu of tools. That’s what Cloudflare is built for.
A chat wrapper answers questions. An agent reasons, then writes and runs the code that gets the job done.
Every run gets its own lightweight execution environment, spun up in milliseconds instead of a server you provision and keep warm.
Answers pulled from your own docs, not memory. State that two requests never disagree on.
A human approval gate, a Workflow that survives a crash, one gateway that routes, rate-limits and audits every model call.
Never touched Workers AI or shipped an agent on Cloudflare before? Good, that’s exactly who today is for.
In under 90 minutes, your team ships something that actually solves a problem in your industry: deployed, working, and yours to keep.
Ease of use and fast iteration matter, but so do security and a human in the loop, whether you run a global enterprise or a five-person team.
Enterprise and digital-native teams, side by side, most touching Workers AI or a Cloudflare agent for the first time today.
You’re not building alone. Five Cloudflare agents already run this event: planning builds, pairing with teams mid-code, and judging every deploy before a human signs off. Meet the crew.
Turns your industry and pain point into a scoped build spec and a paste-ready prompt, in about the time it takes to grab a coffee.
Agents reason about your domain.
Meet Scoping →The pair programmerLives in a chat dock on every page, remembers your team’s context, and pages a human the second it’s actually stuck.
Agents need grounded knowledge and a human fallback.
Meet Mentor →The toolboxExposes the whole event as tools your own coding agent can call directly, no copy-pasting required.
Agents act on your systems.
Meet MCP Server →The refereeA durable Workflow that validates, screenshots and scores your deploy, then waits for a human to actually approve it before it counts.
Durable execution and governance.
Meet Judge →The closerTakes what you shipped and turns it into a draft internal pitch, emailed to you before you’ve packed up your laptop.
The business outcome.
Meet Bring-It-Home →8 tracks: 7 pre-assigned, 1 open.
Delivery-exception agent: detect, decide, compensate, notify.
Pre-assigned podDispute triage with a mandatory human approval gate.
Pre-assigned podTelemetry anomaly, diagnose, book service.
Pre-assigned podAccess-anomaly triage, or heat-pump fleet optimisation on a schedule.
Pre-assigned podIncident triage across runbooks, grounded with AI Search.
Pre-assigned podRights and metadata research agent using Browser Rendering.
Pre-assigned podExperiment-analysis agent that designs its own follow-up test.
Open trackOpen track. Bring your own pain point, or pick one of three suggested shapes.
T-5 days. Link to /hack/preflight. Each team confirms one member can install wrangler.
Hosts catch locked-down laptops before the talk starts.
Opens with a live four-minute build and deploy of agents-starter.
Each team runs the Scoping agent: spec, paste-ready prompt, worker name.
Countdown on /hack/build. Mentor dock live. Hosts work the escalation queue.
Rolling. Form, or submit_project via MCP. Judge Workflow fires per submission.
Eight teams, 2.5 minutes each, on /mission-control. Live human approval.
One cross-industry scenario fanned across registered team agents.
Bring-It-Home agent drafts a personalised internal pitch, emailed.
Every team's project gets scored and approved on stage, then appears on the live board when showcase opens.
Open Mission Control