Summary
Chris Noring, an AI engineering specialist at Microsoft, delivers a conference talk at AI Engineer about how the software developer role is structurally shifting from writing code to designing systems. His central claim: the editor is no longer the natural starting point for development. Instead, he proposes a CLI-first workflow — starting in GitHub Copilot CLI or Claude Code, then moving to the editor for refinement, then using agents (via CLI or GitHub UI) for scaling — a pattern he argues suits the reality that AI now generates most of the initial code.
The talk introduces two concrete tools for agent reliability. First, AGENTS.md files: behavioral contracts placed in a codebase that tell agents how to operate within that project, and which can be copied across projects to adapt new codebases to established conventions. Second, skills: self-contained, strictly constrained instruction sets that agents can invoke for repeatable tasks, stored in `.claude/skills` (for Claude Code users) or equivalent directories in Copilot. Noring emphasizes that all major vendors — Anthropic, Microsoft, and others — have converged on this skills pattern, meaning the investment transfers across tools.
A recurring tension in the talk is the tradeoff between AI-generated code velocity and long-term maintainability. Noring notes that while some organizations are eliminating code reviews to maximize throughput, Glean (and by implication Microsoft) maintains human review as a quality gate — accepting the throughput cost in exchange for understanding the code that gets committed. He frames the developer’s new value as systems thinking and constraint design rather than implementation.
📺 Source: AI Engineer · Published July 11, 2026
🏷️ Format: Keynote Launch
