Keynote: Exoskeletons, not Autopilots
summary
Argues for building augmenting tooling not replacements and gives concrete blueprints: use Elixir + Ash Framework, prefer a monorepo, run agent sessions (Anthropic Claude) via GitHub Actions or an SDK, spin VMs for controlled runs, and log all agent messages; ship a code review bot and Slack/Linear bots that open PRs and QA with Playwright. Results claim: ~2,000 monthly changes, 4 deploys/day, <1% rollback, 5 hours idea→ship, ~6x output in nine months; add proactive nightly PRs to keep docs current.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:33Exoskeletons vs autopilots in AI tooling
- 9:14Core principles for agentic tooling
- 12:02Phase one: Writing code with AI agents
- 17:31Refining agent instructions
- 23:05Phase two: Building and deploying AI bots
- 31:06Internal AI-powered home base
- 35:23Phase four: Integrating agents with workflows
- 40:00Phase five: Making tooling proactive
- 48:06Scaling beyond GitHub Actions
- 51:19Results and business impact
- 52:40Q&A
related talks
54mKeynote: AI-Powered App Development
31mKeynote: How Software Testing Can Increase Agent Autonomy
32mKeynote: Architecting Systems for Agent Experience (AX)
37mAI-Driven Developer Productivity at Scale
24mIt's Time To Rethink Everything
30mLegacy Rails and the AI That Couldn't
15mKeynote: The Self Driving Company
55mKeynote: Who are we Now?