The same arguments from the newsletter, worked through out loud: what a harness actually is, what a "job description" for an agent looks like, and the ten skills that don't get automated away.
11:09
Someone on your team says just call the API. Someone else says you need an MCP server. The same feature built twice, once by hand and once through MCP, so you can see exactly what changes and what doesn't.
Watch on YouTube
12:24
The word everyone in AI throws around and nobody defines. One metaphor and a live build of a real agent, layer by layer, so you can see exactly what each one does.
Watch on YouTube
12:25
Impostor syndrome isn't a skill problem, it's a scoring problem: you keep every piece of criticism and give away every win. AI didn't cause it. It just made it louder.
Watch on YouTube
13:19
25 years building software, and training developers from day one to senior, distilled into the 10 skills AI can't replace: the ones that decide who thrives as the tools get smarter.
Watch on YouTube
14:15
Claude Code gives you an automatic discount on every session, up to 90% off, and most people switch it off without noticing. The six habits that reset it, the ones that keep it on, and the two screens that show you whether any of it is actually working.
Watch on YouTubeSix short breakdowns, one per layer of the harness explained in the video above — the agent definition, the skill, MCP, the context window, governance, and observability.
