AI Courses Are Better as Agent-Runnable Runbooks

·https://vercel.com/academy/slack-agents
articleaiagentsslackvercelai-sdkoperationsobservabilityrunbooksagent-discovery

Vercel Academy's markdown, search, and skills surfaces map directly to joelclaw's agent-discovery layer and runbook-first gateway work.

Demo bots are cheap. This Vercel Academy course frames a Slack agent as a production service: Bolt for JavaScript, Nitro, AI SDK, Vercel deployment, structured logs, retries, fallback behavior, OAuth scopes, SLOs, and an operations runbook that someone else can actually use.

The clever bit is the shape. It doesn’t stop at “put AI in Slack” bullshit. It treats slash commands, shortcuts, modals, App Home, the 3-second acknowledgement rule, Zod tool schemas, AI Gateway, and production logs as one system instead of a pile of disconnected tutorial steps.

The other useful part is the page itself. Vercel Academy exposes a markdown version at /academy/slack-agents.md, an llms.txt, a searchable content API at /academy/search, and agent-facing skills at /academy/SKILLS.md. That’s the real pattern worth stealing for joelclaw: make the learning surface readable by humans, then make the same surface executable and searchable by agents.

For the joelclaw gateway, this maps cleanly: acknowledge fast, move slow work out of the interaction path, carry correlation IDs through every step, and write the runbook before the failure shows up at 3am.

Key Ideas

Links

← All discoveries