Tonight, the alarms. Every night stands on its own.
How you do anything is how you do everything.

AI ExchangeBuilding Phoenix's AI community, together. Workuity hosts tonight, and we present Founder Weekend side by side.
99% of Find dinners under 15 seconds.
Each one: who acts, first three checks.
Roll back first. Debug second.
Logs are JSON, no PII
Down means #alerts in 3 minutes
Grok Bot drafts the triage
✓ lib/log.ts · one JSON line, requestId on every response
{"event":"dinners_found","status":200,"ms":8412}
claudeRemoved 3 console.logs that printed ingredients.
Add a JSON logger with Vercel's request id on every response, redact personal data, use it on the API and cron routes, and show me the lines.
vercel.com/docs/logs/runtime
Claude reviews what it wrote.
The wizard creates the project and config files.
No PII, a sample rate, the requestId tag.
A test error with a readable stack trace.
I run the Sentry wizard. Claude reviews its files, strips PII, tags the requestId, monitors both crons, and proves it with one test error.
docs.sentry.io/platforms/javascript/guides/nextjs
{"status":"ok","db":{"ok":true,"ms":38},
"aiGateway":{"ok":true,"ms":112},"commit":"8c1e2f4"}
claudeDatabase stopped: 503 in 2.1s. Restored.
A /api/health route that checks Neon and AI Gateway without spending tokens, returns 503 on failure, and a Sentry Uptime monitor I set up.
docs.sentry.io/product/uptime-monitoring
connections · 14 active, 31 idle, max 104
slowest · favorites by user_id · 41 ms × 3,120
claudeNo index on favorites.user_id. Want a migration?
I enable pg_stat_statements. Claude writes read-only health queries, a Database section for the runbook, and checks we use the pooler.
neon.com/docs/introduction/monitoring
Every alert has an owner
and a first check.
99% of Find dinners under 15 seconds.
App down: three failed checks in a row.
One 429. One retry. One bad webhook.
Interview me, then extend RUNBOOK.md: on call, SLOs, alerts with owners, first checks, rollback, logging, the Grok Bot routine and a postmortem template. I create the Sentry rules.
docs.sentry.io/organization/integrations/notification-incidents/slack
youFind dinners is slow since 2pm.
claudeSentry: p95 31s since release 8c1e2f4. Logs: 3 retries per request.
claudeRoll back or fix forward? Your call.
I add the Sentry and Vercel MCP servers. Claude investigates an alert read-only across Sentry, Vercel logs and workflows, then writes a triage note.
mcp.sentry.dev
Sentry · Error spike · POST /api/dinners
claudeHealth ok. Started 14:02, after commit 8c1e2f4.
→ Triage note and a Claude Code prompt drafted. Approve to post?
Write a Grok Bot routine brief: on each #alerts message it reads the alert and /api/health, drafts a triage note and a Claude Code prompt, and asks before posting.
docs.x.ai/grok-bot
✗ Sentry · TypeError in POST /api/dinners (preview)
→ #alerts · Grok Bot draft in the thread
claudePostmortem written. 2 actions, both owned.
Break a preview on purpose, time the path from error to #alerts to Grok Bot's draft, rehearse the rollback, and write the postmortem.
Also in the series: Part 11, Admin Console. The controls to act on what monitoring finds.
feature:dinners · 61% of spend
feature:photo · 33% · up 4x this week
claudePhoto spend jumped after Tuesday's deploy.
Tag AI calls by feature and user, set a budget with spend alerts, and print a weekly spend-by-feature table.
vercel.com/docs/ai-gateway/observability-and-spend
An Arizona AI hackathon. Teams of 1-4.
See the hackathon ↗