Agentic Coding at scale at Open Mercato
summary
Open Mercato runs an agentic software factory that produces about 1,200,000 lines of code, ships biweekly, and coordinates 140+ contributors. It replaces a single AGENTS.md with nested AGENTS.md routing, moves to spec-driven development (markdown specs reviewed by multiple models), and encapsulates common work in reusable MIT-licensed skills (code review, setup, ephemeral test envs). Orchestration uses Caesar (run with npx caesar-cli), heavy GitHub CI, and automated PR QA; a 300-PR hackathon proves the system scales.
AI-written summary · about
chapters (22)
- 0:00Introduction to Open Mercato
- 1:12A platform built by agents
- 2:34What a software factory means
- 3:02From single-shot prompting to a system
- 4:32When one AGENTS.md file stops scaling
- 5:08Task routing with nested instructions
- 5:44Spec-driven development
- 7:37Skills as factory machines
- 9:03Open-source skills
- 9:28A 300-PR hackathon
- 10:20Skills across the software lifecycle
- 11:20Ephemeral test environments
- 11:48GitHub as the coordination surface
- 12:40Standardizing repositories with a setup skill
- 13:18Scaling concurrent work
- 13:42Orchestrating hundreds of tasks with Caesar
- 15:30CI as the safety net
- 16:06Separating core and user context
- 17:11Compiling dynamic context
- 18:08Evals from PR and session data
- 19:07Cloud sandboxes
- 19:40Closing lessons
related talks
35mAgentic coding
25mBeyond the IDE: Toward Multi-Agent Orchestration
40mHow To Tame Your Agents?
01h 28mAI Coding Agents and how to code them
24mAgentic Al in SW Development: Evolving Patterns & Protocols
01h 05mPanel: Agentic SE & Open Source in the FMware Era – Toward Trustworthy, Context-Aware FMware
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
25mFrontiers of Agentic AI