The manuscript for Life in Perspective is in production. November 3 is locked in as the release date, and we're now heads-down on the launch.

First up: pre-order bonuses. I'm putting together something worth your time, and I'll share details soon.

If you've already pre-ordered, thank you. You'll be able to claim the bonuses when we announce them.

If you haven't yet, now is a good moment. Pre-orders tell publishers (and the world) that this topic has an audience. It makes a real difference.

Inside Claude Cowork: Watch AI Execute 7 Complex Tasks

You've done this routine a thousand times. Open a chat. Find the file. Drag it in. Wait. Then find the next one and do it again. That’s over. Claude can read the files already on your computer.

In this video, I walk through what file system access changes. Claude Cowork reads whole folders where they live. It edits files in place. It drops new docs into the right Project folder. And it can pull from two folders at once to surface contradictions and gaps you'd never spot one file at a time.

If you have a Second Brain, this is the feature that makes your file system the center of the work again.

Plan First, Then Restart Before You Build

Most project managers will tell you: keep planning and execution separate. I've found the same principle applies to working with AI.

Worth noting upfront: this is for when you're trying to produce something concrete. A document, a proposal, a plan, a presentation. Not a quick question or a casual back-and-forth.

When the goal is a real output, I now treat the first session as planning only. I work through what I want to build, what I need, what approach makes sense. Then, before building anything, I stop.

I copy what we've created together (the outline, the brief, the spec) and open a brand new session.

A few reasons this is worth the extra step:

• Long, meandering sessions accumulate irrelevant context, and performance can degrade because of it.
• Stopping before you build creates a natural moment to ask: do I actually have everything I need?
• If the session crashes or glitches midway, you have a recovery document ready to go.

Try it this week: next time you're working toward a concrete AI output, treat the first session as planning only. When you're ready to build, start fresh and paste in your brief.

What do you notice? Hit reply and let me know.

How Meta Built an AI Second Brain for 60k Knowledge Workers

The Analytics team at Meta recently shared a breakdown of how they built an internal AI Second Brain. One finding stood out to me: feeding an AI agent everything at once actually makes it worse.

Their solution was what they call progressive disclosure. Start with lean context, drill into the details only when the conversation needs it.

PARA turned out to be the perfect structure for this, because it already sorts information by what's active versus what's just stored.

One data scientist built it. No top-down mandate. 63,000 employees adopted it in three months.