Mike Taylor

<start of post>

I’ve spent the last 48 hours trying to break the new Claude model, and I think I finally found its ceiling.

Most people are using it for basic coding tasks, but I wanted to see if it could handle high-stakes strategic planning. I fed it three years of raw financial data and asked it to find the "hidden leaks" in our customer acquisition cost.

It didn't just find the leaks; it built a custom attribution model that accounted for seasonal decay. I actually enjoy reading the logic it uses to justify its decisions. It feels less like a calculator and more like a senior analyst who’s had too much coffee.

I also tested it on my "impossible" benchmark: rewriting a 2,000-word technical essay into the style of a 1920s noir novelist without losing the technical accuracy. It nailed it on the first prompt.

The hard part now isn't the tool—it's the imagination of the user. If you're just asking it to summarize emails, you're leaving 90% of the value on the table.

I put together a full breakdown of the prompts I used and the raw outputs over on the Every site. Give these a try on your most ambitious project this week:

<end of post>