The Copilot Connection Ep 37 - Coding Agents
Jan 30, 2026
Garry Trinder, a Microsoft developer advocate for Microsoft 365 Copilot, explains coding agents and how they reshape development workflows. Short takes cover agent interfaces, context and memory, spec-driven vs exploratory approaches, tooling and skills for repeatable automation, effects on junior developers, personality in agents, and security and ethical guardrails.
AI Snips
Chapters
Transcript
Episode notes
From Editor To Agent-Led Development
- Coding agents shift the developer from editing code to instructing an assistant that can read, modify and run the project autonomously.
- Agents combine model reasoning with access to tools (files, terminal, APIs) to iterate and fix code without constant user triggers.
Preload Project Instructions
- Provide project-specific instructions (style, folder structure, tech choices) so the agent generates consistent and expected code.
- Store those preferences in an instructions file the agent can load to reduce non-deterministic outputs.
Make Agent Memory Explicit
- Memory and decision logs let agents recall prior design choices and reduce repeated discussions across chats.
- Persisting decisions (e.g., markdown decision logs or git history) yields better continuity and personalization for future work.