Agentic Elixir
summary
Agents start sessions by reading the repo so onboarding costs compound — the team records 648 sessions in four months and uses an agents.md entry point plus markdown skills (Sentry CLI, Linear CLI, LaunchDarkly patterns) to speed context loading. Deterministic tools like mix format and Credo checks reduce agent errors; agents write Credo rules to harden conventions. Webhooks into Claude produce about a 60% success rate for Sentry auto-fixes. Worktrees, cached build artifacts and cutting a 10s task startup to ~1s make agent loops practical.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:06Why we use agents at Knock
- 3:09Agentic development pillars
- 5:59Efficient agent onboarding
- 12:14Empowering agents with tools
- 18:52Optimizing human workflow
- 20:41Key takeaways
- 21:03Q&A
related talks
32mWhat Agents Do When You’re Not Looking
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
49mAgile for Agents Managing Robots The Way We Manage Humans
07h 47mAI Agents Conference 2026 - Day 5
33mAgent Experience
21mThe Agent Native Company
20mThe Pod: Agents Are Developers Too: The Helius Agentic Stack
40mHow To Tame Your Agents?