We asked Copilot and Cowork to forecast August 2026 US flight delays and cancellations, then checked the predictions against what actually happened. A simple statistical average beat them all.

We turned an AI policy into a five-minute interactive training experience. The included Microsoft 365 Copilot won, while Cowork Max built something beautiful that we could not share.

One short prompt, a 90-second animated explainer with narration and captions. Copilot struggled, Cowork Light stumbled, and Cowork Medium and above delivered results that genuinely surprised us.

We asked Copilot Chat, Microsoft 365 Copilot, Copilot in PowerPoint, and Cowork to turn our website into a 12-slide customer deck with speaker notes. The free option surprised us, and so did the paid one.

Can Copilot vibe code a working insurance quotation app from a 191-page manual? We tested Copilot Chat, M365 Copilot, Researcher, Cowork, and Claude. The results varied dramatically.

We tested every Cowork effort setting, from Light to Max, building an Excel executive dashboard from Canadian wildfire data. The most surprising result: a polished dashboard that reached the wrong conclusion.

We gave four Copilot options a real nine-document RFP for a Cisco-to-Teams Phone migration and asked for a bid recommendation. Only one caught the risks that really mattered.

In the first episode of Calibrating Copilot, we run the same project-summary prompt through Copilot Chat, M365 Copilot, Researcher, and Cowork, and rate each on accuracy, completeness, speed, and cost.