AI DevOps Podcast Michael Perry: AI-assisted Development - Episode 397
22 snips
Apr 13, 2026 Michael Perry, Director at Improving Enterprises and software mathematician who builds immutable architecture and created the Jinaga runtime. He discusses first-principles thinking for AI, which tasks AI can and cannot automate, multi-agent workflows and RooCode subagents, test-driven AI development, and turning team practices into sharable software factory assets.
AI Snips
Chapters
Books
Transcript
Episode notes
Why One Engineer's Preference Hooked Michael Perry
- Michael Perry got hooked on software when an incomplete app he rewrote worked better for an engineer doing real work than the production app it replaced.
- That moment convinced him the work of programming amplifies other people's impact and kept him fascinated by the craft.
Separate Automatable Tasks From Human Judgment
- Jobs decompose into automatable tasks and high-value human tasks like taste, discernment, and building trust.
- Michael Perry argues humans should offload compressible work to AI and focus their uniquely human judgment and relationships.
Make LLMs Validate Their Own Code
- Close the loop: have the LLM write tests or drive the UI so it sees test results and can iterate when code fails.
- Use TDD or UI automation (Playwright, macOS automation) so the agent can detect failures and correct itself.
