-
The Context Harness and the Execution Harness: Two Harnesses, One Loop
September 25, 2026
The context harness holds what your agent knows. The execution harness does the work. Why 80% of my agentic time goes into the first one.
Read More → -
A Data Agent Should Be a Sub-Agent
September 23, 2026
A Fabric data agent works best as a sub-agent with one job that Copilot, Foundry, or Copilot Studio calls. Here's how I write those instructions.
Read More → -
A skill.md Is Not Wisdom. So Make the Agent Grill You.
September 21, 2026
Your skill file carries instructions, never the judgment that says this is six hours not two. Matt Pocock's grill-me is how I get the judgment back in.
Read More → -
Central BI Is a Tiger Team Now
September 18, 2026
In Fabric, central BI shouldn't own workspaces and ship gold models forever. It should float to the priority like a tiger team. Here's the operating model.
Read More → -
Build the Thing That Creates the Thing
September 16, 2026
Stop prompting for the same notebook edit. Build the executor: an agent, its skills, and the context it reads, in a harness that reaches Fabric and Power BI.
Read More → -
Don't Hire a Senior to QA the Agent
September 14, 2026
Hand agents to people who can't interrogate the output and you haven't scaled your team. You've built a review queue and staffed it with your architect.
Read More → -
You Don't Have an AI Strategy. You Have AI Theater.
September 9, 2026
The intake test I run on every new skill, MCP server, and harness before it touches client work.
Read More → -
Governance Pillars 3 and 4: Write the Contract Before You Hire the Watchdog
September 2, 2026
Accountability and culture are where governance gets hard. The data contract that gives an agent something real to enforce, plus a 30-day plan.
Read More →