AI DevOps Podcast Daniel Ward: AI Agents - Episode 393
11 snips
Mar 16, 2026 Daniel Ward, a Microsoft .NET MVP and consultant focused on CI/CD, testing, AI adoption, and product management. He explains AI agents as LLM-driven actors that take actions, shows practical patterns for verify-and-iterate workflows, contrasts Copilot SDK orchestration with inline prompts, and explores agent use cases like PR analyzers, log-driven fixes, and personal research agents.
AI Snips
Chapters
Transcript
Episode notes
Agent Means Autonomous LLM Decision Making
- Daniel defines an AI agent as an LLM that can make autonomous decisions like deciding to browse the web instead of only responding to prompts.
- Agentic behavior means the model orchestrates actions beyond single-turn prompts, e.g., fetching data or running tools on its own.
Make Agents Run Deterministic Build And Test Scripts
- Embed deterministic scripts (build, run, tests) into your Copilot instructions so the agent iterates until the change works.
- Use Copilot's persistent premium requests to let it loop for minutes or hours until the code builds and runs correctly.
Copilot SDK Lets Code Orchestrate Prompts
- The Copilot SDK flips orchestration: instead of prompts driving code, deterministic C# code drives prompts and calls into Copilot.
- This lets you embed business logic and UI handling in code while using Copilot for analysis or generation tasks.