Humans in the Loop
Learning AI in public

Learning AI one prompt at a time

Bruce Cullen builds with AI tools during the working week and reports back what held up — agents, products, governance, and the week’s news.

Listen to EP 004All episodes
Humans in the Loop
EP 004
What a computer-using agent actually does all day
Agents · Governance · NewsBruce Cullen
Latest episodes

One conversation a week

All episodes
EP 004

What a computer-using agent actually does all day

Two weeks of screen recordings, one honest tally: where the agent earned its keep and where it quietly wasted an afternoon.

AgentsGovernanceNews12 Aug 2026 · 38 min
EP 003

Prompting is a management skill now

Why the people who delegate well to juniors delegate well to models, and what that means for hiring.

PracticeTeams05 Aug 2026 · 41 min
EP 002

The evaluation problem nobody wants to own

Everyone agrees you need evals. Nobody agrees whose job it is. We work through a real one, badly, then better.

EvalsGovernance29 Jul 2026 · 35 min
The model was fine. The workflow around it was the problem.
Bruce CullenEP 004
Why the show exists

Most AI advice is written by people who do not have to ship anything. This one is a running log of what worked in a real week of work, and what quietly did not.

About the show
From the blog

Notes between episodes

All posts
23 Feb 2026 · 2 min read

The cost of a second opinion

Running a second model over the first model's output caught real mistakes. It also doubled the bill and added a step nobody owned.

AgentsGovernance
09 Feb 2026 · 1 min read

Why this show exists

Most AI advice is written by people who never have to ship anything. This is the other version: one week of real work, reported back.

MetaProducts

Follow the show

A new episode every week, wherever you already listen.

RSS