Yesterday I sat across from a legendary investor: one of those people who’s been “right” more often than seems statistically possible.

No special dashboards on the wall.

No elaborate note-taking ritual.

Just a morning routine that starts before sunrise, a constant stream of incoming information, and an almost eerie ability to notice when something in the world stops behaving “the way it normally does.”

At one point, they said they get two to three thousand emails a day, but only deeply read a small fraction. The rest are skimmed, mentally triaged, and deleted.

That single detail is the most common pattern I see across tech CEOs, fund leaders, PE partners, and nonprofit executives: an overload of inputs competing for the same scarce resource (attention).

They’re already using modern LLMs like a high-end research analyst: quick questions, rapid synthesis, sometimes letting the model “think” and printing the result to read later. And they’ve noticed what we’ve all noticed: the models are dramatically better than even a few months ago.

But the most interesting part of the conversation wasn’t “which model is best.”

It was the operating system underneath their decision-making. Their worldview is built as a living matrix across a handful of interacting buckets (think: major asset classes / major levers), informed less by headline macro data and more by on-the-ground signals (e.g., company-level information, pattern breaks, unusual flows, the subtle stuff).

When something violates the pattern, that triggers action: a call to a trusted contact, a deeper dive, or an LLM query.

That’s the lesson for organizations adopting AI: don’t start with “AI, what should I do?”

Start with: What are my triggers? What do I consider a pattern break? What inputs do I trust and which ones are just noise? Then AI becomes powerful in a very specific way: not replacing judgment, but tightening the loop:

• A daily brief that prioritizes the top 15 items that genuinely matter.
• An inbox that highlights signal instead of volume.
• A repeatable “document digestion” workflow so you’re not rewriting prompts every time.
• A second model that challenges the first, so you get debate instead of a single confident-sounding answer.

AI doesn’t win by being smart. It wins when it’s wired into how you already make decisions, without drowning you in more words.

These days, with Claude, it is easier than ever before. But it's not without cost:

• It requires connections to the software that holds the data that informs your decisions.
• It requires an investment, by you, in the platform. You need to train it. You need to tell it. This is not $ but time.

Alex