← back
Making Codebases Agent Ready – Eno Reyes, Factory AI
Takeaway
The bottleneck for agent productivity in enterprises is automated-validation rigor in the codebase, not which coding agent you buy.
Summary
- Reyes (Factory AI) argues software is the frontier for agents because it's highly verifiable; the agent ceiling is set by how well you can specify objectives and validate outputs.
- Lists 8 pillars of agent-readiness: opinionated linters, high-coverage tests, non-flaky builds, OpenAPI specs, AGENTS.md docs, format validators, etc.; humans tolerate 50-60% test coverage but agents break without rigor.
- Pushes specification-driven development (spec mode / plan mode in Droid, Cursor) — specify constraints, generate, verify, iterate — over the traditional understand-design-code-test loop.
- Argues organizations should invest in validation infrastructure rather than spending 45 days comparing coding tools on SWE-bench; this unlocks parallel agents, large-scale modernization, and AI code review.
agentscode-generationverification
Original description
Agents are eating software engineering. Yet teams deploying these tools face mixed results. Agents work great in demos but fail unreliably in production, frustrating engineering teams who expected better. The gap isn't model quality—it's environment readiness. Agents need fast feedback loops, explicit instructions, and predictable environments to work effectively. They break on missing environment variables, undocumented dependencies, and tribal knowledge that "everyone just knows." What if you could measure and fix what's holding your agents back? Enter Agent Readiness. In this talk, we'll explore eight categories that determine whether your codebase is agent-ready: from style validation and build systems to dev environments and observability. You'll learn how to score your repos, identify easy wins, and build environments where agents actually ship reliable code. We'll share real signals from Factory's work running autonomous agents in enterprise production repos—and give you a practical framework to make your team's agents more productive starting tomorrow. Speaker: Eno Reyes | CTO, Factory AI https://x.com/EnoReyes https://www.linkedin.com/in/enoreyes/ https://enoreyes.com/ The video argues that the primary bottleneck for adopting AI agents in software engineering is not model capability, but rather the "agent readiness" of the codebase—specifically the rigour of automated verification systems. Eno Reyes from Factory AI posits that software development is shifting from a specification-based process to a verification-based one (Software 2.0), where the ability to mechanically validate code (via linters, tests, and strict environments) determines an agent's success. He suggests that organizations must invest in these feedback loops to create a "flywheel" effect: better environments lead to better agents, which in turn free up time to further improve the environment. 00:00 Introduction & Factory AI Mission 01:19 Software 2.0: Automation via Verification 02:21 The Asymmetry of Verification (P vs NP) 04:01 Automated Validation as an Agent Constraint 06:09 Shift to Specification-Driven Development 11:51 The New DevX Loop: Investing in Feedback Cycles 13:42 Conclusion: The ROI of Agent Readiness