5 posts tagged with "Agent"

In a one-day AI workshop for non-developers, initial setup alone took 2 hours and we only got through half the curriculum. Here's how I changed the way I run workshops, in 5 steps, including a check script participants run the night before and a timetable for the day.

If information can't leave the company, it doesn't go into AI either. Here's a 5-step process: checking ChatGPT and Claude training settings, a 4-line do-not-enter list, a Python script that masks and restores names, amounts and contact details, and a review checklist prompt.

Before you use an AI-generated analysis as is, check the baseline period, the sample, the unit being counted, source bias and causal order. I show how with calculations on real public data, plus a review prompt you can paste into ChatGPT or Claude and a 15-item checklist.

We hooked six models up to the dinosaur game and compared them over two rounds, looking at real API cost and survival time. Then we gave the same states to 7 models, including Solar, and compared decision accuracy, API response time and actual call cost side by side.

We fetch the same weather data twice: once by calling the API directly, and once by wrapping it in an MCP server and opening the tool description the AI actually reads. With a free public API and a single Python file, you can see the difference between API and MCP for yourself in 15 minutes and decide when to use which.