Paul Duvall, author of Continuous Integration, argues that agentic AI is reinforcing rather than replacing core engineering discipline. As AI generates increasing volumes of code, practices like trunk-based development, automated testing, and specification-driven development become more critical. Duvall's open-source repository documents evolving patterns for AI-assisted development, including spec-driven workflows, codified rules, and observable development. Paul Stack and Gergely Orosz echo similar themes: Stack restructures workflows around GitHub Issues and agent-readable specs instead of pull requests, while Orosz describes 'remixing' contributed PRs via agents. All three emphasize that clear upfront specifications, architectural constraints, and shift-left/shift-right feedback loops are essential to avoid unpredictable AI output. Duvall also suggests AI may enable smaller, more focused teams as coordination overhead decreases.

5m read timeFrom infoq.com
Post cover image

Sort: