Articles on agentic AI in software engineering.


Most public writing about agentic design patterns either jumps to deployed pipelines or stays inside the model harness. I take a look at the layer most software engineers are actually working at.

A walkthrough of an orchestration pattern for agentic coding: a single orchestrator dispatches scoped subagents to independent tasks, and an independent reviewer inspects each result. The examples use Claude Code, but the concepts apply to any modern agentic coding tool.

Ticket writing and pull request workflows aren't valuable because they're convenient for humans; they're actually quite inconvenient upfront. Their value comes from something less visible: they determine what engineering teams are able to notice, catch, and correct.

Software engineering is in a middle period of AI integration. Four open questions define the next few years: which systems to evolve, which to leave behind, where engineers actually work, and how teams keep trust intact.