Weekly deep dives on context engineering, agent architecture, and building software that builds itself.
Senior Software Engineer @ Microsoft (GitHub · Copilot Ecosystem)
I build agentic systems — software that doesn't just assist developers, it collaborates with them. My work lives at the intersection of context engineering, agent governance, and platform architecture.
By day, I work on the GitHub Copilot ecosystem at Microsoft. By night, I run a 43-agent AI home assistant that manages my family's entire life — calendars, meals, finances, health tracking, content creation, and more — all powered by GitHub Copilot CLI.
My core thesis: Make the right thing to do the easy thing to do — for humans AND agents.
I write about what I learn, ship the tools I build, and teach the patterns that actually work. If you're building with AI agents, you're in the right place.
I create content at the frontier of agentic software development. Here's what I write and teach about:
| Pillar | What It Covers |
|---|---|
| 🧠 Context Engineering | Prompt architecture, memory systems, skill files, and making AI agents actually useful |
| 🤖 Agent Skills & Architecture | Multi-agent orchestration, delegation patterns, agent-to-agent communication |
| 🔌 MCP Ecosystem | Model Context Protocol servers, tools, and integrations — including turning your phone into one |
| 🪝 Extension Architecture | Copilot hooks, extensions, governance layers, and guardrails for safe AI coding |
| 🔒 AI Governance | Safe agentic development, content signing, approval workflows, human-in-the-loop patterns |
| ⚙️ Platform Engineering | GitOps for AI, infrastructure-as-code for agent systems, CI/CD with AI in the loop |
📬 Get the deep dives in your inbox → Subscribe to the Newsletter
- [When GitHub Copilot Extensions Go Wrong — Part 1](https://htek.dev/articles/stop-building-fat-copilot-extensions/) — One unclosed async handler took down all 43 of my Copilot agents. Here's what I discovered about extension failure modes and an idea I've been cooking up: the hollow extension pattern.- [Stop Connecting Your Agents One by One](https://htek.dev/articles/stop-connecting-your-agents-one-by-one/) — Every AI harness has a different bias. Wiring them one at a time is outdated. Here's the messaging layer I built — MeshWire public beta is now open.- [I'm Hunting for My Vertical](https://htek.dev/articles/im-hunting-for-my-vertical/) — I spent a week building AI across five industries and realized I've been thinking about this all wrong. Now I'm hunting.- [You're Not Doing GitOps (You're Doing CI/CD With Extra Steps)](https://htek.dev/articles/youre-not-doing-gitops-cicd-with-extra-steps/) — Most teams think they're doing GitOps because deployments trigger from Git. They're wrong. Here's the enforcement pattern that makes the difference.- [Your GitHub Actions Don't Need Secrets](https://htek.dev/articles/your-github-actions-dont-need-secrets/) — How a custom OIDC broker, reusable workflows, and layered frameworks turned GitHub Actions into a self-service platform for 300 engineering teams.
Languages
AI & Agents
Cloud & DevOps
Frontend & Frameworks
