When I heard that Microsoft Copilot could now run on Anthropic’s Claude, I was curious. Claude’s smarts plus Copilot’s access to all of CGNET’s documents and email? That sounded like a pretty good combination.

So I tried it. My verdict: so-so.

When AI Has More Information Than It Knows What to Do With

I’d been using Claude and ChatGPT to help write proposals for a while and poking at Copilot on the side. Here’s what surprised me: more information didn’t get me better answers. Copilot can dig through a mountain of company files, and sometimes it seemed bent on using all of them. Old proposals, unrelated email threads, documents that barely touched the project — in they went.

Most of it was accurate.

Not much of it was useful.

When Context Beats Access

Claude and ChatGPT were usually more on point. After working on proposals with me for a while, they’d picked up a decent sense of CGNET, our clients, and how I think through a business problem. They didn’t know everything about the company. They just had a better feel for what mattered.

Here’s an example. We were preparing a contract that named two people in it: Aaron as the signatory and Marianne as the project manager. Claude questioned whether we had the right project manager. It remembered that Aaron had introduced me to Marianne just a few weeks earlier and suggested that James might be the better person to send it to. Claude wasn’t necessarily right. But it got my attention — and floored a colleague who happened to be watching — because it connected an earlier conversation to the task at hand and used that context to challenge my thinking. That’s not search. That’s deciding what’s worth bringing up.

Which is exactly why the idea of Claude inside Copilot interests me.

Knowing What to Ignore

We’ve spent decades assuming that the more a computer can see, the more useful it is. Enterprise AI takes that idea about as far as it can go. But think about a good executive assistant: They’re not valuable because they’ve memorized every file in the building. They’re valuable because they know what’s a priority, who matters to whom, what’s about to go sideways — and what you can safely ignore.

That’s the test for Claude in Copilot. Plugging Claude into all our company data could be a big deal. But handing it more documents won’t, by itself, make the answers any better. The question is whether it can tell the difference between what’s available and what’s relevant.

To be fair, I haven’t used Claude inside Copilot enough yet to say how it’ll do. That’s why I plan to keep testing it.

The Buffet Problem

So…about that title: An all-you-can-eat buffet might give you fifty dishes, but more choices don’t necessarily make for a better dinner. Sometimes I’d rather have a cook who knows what I like and makes exactly that. Enterprise AI may be headed in the same direction. The next big step may not be giving AI everything an organization knows, but instead teaching it to identify the slice that matters to the person asking.

The smartest AI won’t be the one that piles everything onto the plate. It’ll be the one that knows what matters — and leaves the rest behind.

Want to learn more? AI has been a subject of my writing for several years, and CGNET has offered AI user training and implementation for both large and small scale organizations.   I would love to answer your questions! Please check out our website or drop me a line at g.*******@***et.com.