Demo: Five-minute rig walkthrough

Narration script. Timing in brackets. Future voiceover follows this script verbatim.

[0:00] Opening beat

You file a two-line issue. The rig ships a website. No other human touches anything between those two events.

Here’s what that looks like in real time.

[Screen shows: dashecorp/infra GitHub — Issues tab. Cursor clicks “New issue.”]

[0:20] Filing the issue

The issue is about as minimal as it gets: “New site: carrots.dashecorp.com.” We add the agent-ready label and submit.

[Screen shows: issue submitted. Issue number #119.]

Within two seconds, rig-conductor’s webhook fires. The conductor parses the label, classifies the issue as infrastructure work, and dispatches it to Dev-E-node.

[Screen shows: conductor dashboard — Issues tab. Issue #119 transitions from “open” to “assigned.” Agent column shows “dev-e-node.”]

[0:50] Dev-E works

Dev-E has the issue. She fetches the rig brain — the 27 KB compiled system state that tells her: which repos exist, which Terraform patterns are already in use, what the Cloudflare Pages naming convention is.

She knows the pattern. The rig already has a for_each map for Pages projects, DNS records, and GitHub repos. She adds one entry to each.

[Screen shows: GitHub — PR infra#120 opens. Files changed: 3. cloudflare/pages.tf, cloudflare/dns.tf, github/repos.tf. All adding one line to a for_each map.]

Five minutes from issue to PR.

[1:20] Review-E’s cron fires

Every five minutes, Review-E wakes up and scans for open PRs authored by Dev-E. She finds infra#120.

She reads the diff. She reads the AGENTS.md for the infra repo. She checks the issue context. Then she posts her review.

[Screen shows: Review-E’s review comment appears on infra#120.]

“Approved. Terraform resources correct. DNS record uses proxied: false — confirm this is intentional for a Pages CNAME.”

The operator confirms. Merged. OpenTofu applies.

[Screen shows: PR merged. Terraform apply in CI: 3 added, 0 changed, 0 destroyed.]

[2:00] The failure

Here’s where it gets interesting. The site comes up — but the domain returns a Cloudflare error 1014.

[Screen shows: browser — carrots.dashecorp.com — CF error page.]

Error 1014 is a documented Cloudflare edge case: the Pages project name carrots is stuck in CF’s internal binding registry, probably from a prior account. No Terraform change can fix a stuck binding.

Dev-E tries three times. Dependency ordering. Force-recreate. DNS proxy mode. Each PR applies cleanly. Each time, 1014 persists.

[Screen shows: fast-forward through infra#121, #122, #123. Each shows “Applied cleanly” then browser: still 1014.]

Most tools stop here. Or a human gets paged.

[3:00] The real fix

After three failed attempts, Dev-E diagnoses the pattern: every fix preserved the project name. The name is the problem. A fresh name has no prior binding.

[Screen shows: infra#124 diff — one line changed: name = "carrots-site".]

PR infra#124: rename the project to carrots-site. Review-E approves in four minutes. Merged. CF binding completes.

PR infra#125: provision the three secrets the deploy workflow needs. Merged. Deploy runs. Site is live.

[Screen shows: browser — carrots.dashecorp.com resolves. Page title: “Carrots — Orange. Crunchy. Chronically Underrated.”]

Five PRs. Two hours. The only human action: three PR approvals.

[3:40] Dashboard tour

This is the conductor dashboard.

[Screen shows: rig-conductor /dashboard. Costs tab.]

The Costs tab shows what today’s work cost, broken down by agent and repo. The carrots bootstrap: $1.47 across five sessions.

[Switch to Issues tab]

The Issues tab shows the current queue: assigned issues, their age, which agent holds them. Red rows are stuck — no heartbeat for 30+ minutes. The StaleHeartbeatService re-queues them automatically.

[Switch to Agents tab]

The Agents tab shows each pod’s last heartbeat. Three Dev-E variants. Review-E. Green across the board.

[4:20] What this means

The rig isn’t magic. It makes mistakes. Dev-E picked the wrong fix three times before the right one. Review-E occasionally misses context a human reviewer would catch.

What the rig changes: iteration speed. A human engineer diagnosing CF 1014 would spend 20 minutes reading Cloudflare docs before trying the name change. Dev-E tried three things first but got to the right answer in two hours total.

And every failed attempt is documented. The CF 1014 / project-name collision is now in memory. The next agent skips three cycles and goes straight to the rename.

The rig learns.

[4:50] Close

One issue. Five PRs. A live website. No human wrote a line of code.

The rig handles the iteration. You handle the intent.

[Screen returns to the landing page: rig.dashecorp.com]

Docs at research.rig.dashecorp.com. Whitepaper at research.rig.dashecorp.com/whitepapers/the-autonomous-engineering-rig/.

[END]

Production notes

Section	Screen recording needed	Notes
Issue filing	Yes	Record on dashecorp/infra with real issue
Conductor dashboard	Yes	Use `kubectl port-forward` or Cloudflare tunnel
PR review (Review-E)	Yes	Can use any real Review-E approval
CF 1014 error	Screenshot	Browser screenshot of actual 1014 page
Fast-forward PRs	Screen recording	Speed 4x through #121–#123
Final site	Yes	carrots.dashecorp.com or equivalent
Dashboard tour	Yes	Annotate tabs in post-production

Total estimated recording time (raw): 30–45 minutes. Edited to 5 minutes.