Roasting Code for Fun & Profit with Structured AI Workflows
summary
Shopify ships Roast gem, a Ruby CLI and YAML workflow framework that composes prompts, shell commands, custom Ruby steps and MCP coding agents (Claude Code) to grade, generate and patch code and tests. It uses Ray, supports parallel steps, caching and session replay, and offers context compaction to avoid LLM token limits. Internal adoption spreads across dozens of teams and thousands of engineers; a conservative error-rate example shows 0.1% per-turn nondeterminism already drops multi-step agent success rates (20 turns → ~98%), motivating deterministic, structured workflows.
AI-written summary · about
chapters (8)
- 0:00Structured AI workflows
- 6:31What is Roast
- 10:06Roast workflow features
- 13:37Simple Roast demo
- 17:54Advanced workflow steps
- 21:49Built-in tools & context
- 25:43Roast alternatives & roadmap
- 27:27Q&A
related talks
33mMeet Shopify's Ruby and Rails Infrastructure Team
30mUsing Ruby to Automate Your Life
33mUsing Ruby to Automate Your Life
30mAI-assisted coding: practical lessons from small startups...
37mPatterns of Application Development Using AI
35mThe State of the Ruby AI Toolbox
25mDirecting a Swarm of Agents for Fun and Profit
43mBuilding AI agents in Ruby