Agentic AI Engineering Patterns: How Industry Leaders Reinvent Software Development

From Code Review to Agent Orchestration

Leading voices including Paul Duvall, Paul Stack, and Gergely Orosz are charting a new course for AI-assisted development as agents become capable of generating large volumes of code.

The Core Shift

Paul Duvall, author of Continuous Integration, states: When AI can generate code, engineering practices are more important than ever. Traditional code review does not scale when AI produces hundreds of lines per minute.

Key Patterns

1. Specification-Driven Development - Define specs that agents follow with pre-defined behaviors, constraints, and acceptance criteria.
2. No Pull Requests - Paul Stack's System Initiative team abandoned PRs entirely in favor of GitHub Issues with design proposals.
3. Atomic Decomposition - Complex tasks broken into smaller units executed by parallel agents.
4. Observable Development - Production telemetry fed back into development lifecycle.

Future: Smaller Teams

Duvall predicts AI will enable one-pizza teams through reduced coordination costs and increased automation.