Read time: 5 minutes

Leader, welcome back.

Big week. Microsoft made a significant move. OpenAI shipped something notable. And Anthropic published research that should make every org rethink who they're actually hiring for. Let's get into it.

Microsoft just announced Copilot CoWork, a new M365 feature built directly on Anthropic's Claude technology.
You describe an outcome, CoWork breaks it into steps and produces decks, briefing docs, and workbooks across your M365 apps in the background.

The key difference from Claude CoWork: it runs in the cloud, not on your desktop.
It pulls from your emails, meetings, files, and Teams chats across M365.

What this means for you practically: if your organisation runs on Microsoft, you may not need to change anything.

No separate app. No new desktop agent to manage.
CoWork lives inside the tools you already open every morning.
Ask it to pull together a briefing from last week's emails and meeting notes, draft a board update from your latest files, or turn a folder of data into a formatted workbook. All without leaving M365.

That's a meaningful unlock for finance leaders who've been interested in CoWork but found the desktop-only setup too frictional.
The catch: it's in limited research preview right now, inside a new $99/user enterprise tier. Not available to everyone yet.
But if you're in a Microsoft environment, this is the one to watch.

OpenAI shipped GPT-5.4 this week and it's a genuine step up.
Better reasoning, stronger coding, and notably improved at navigating desktop tasks, which matters for anyone building or using AI agents.

The practical takeaway: if you're doing serious knowledge work in ChatGPT, make sure you're selecting the most capable model available, not just whatever loads by default. The gap between models is wider than most people realise. Defaulting to the faster, cheaper option costs you on quality.

Researcher Noam Brown's comment post-launch: "We see no wall."
The frontier is still moving fast.

Anthropic published research this week mapping which jobs AI is already touching. Computer programmers and data entry workers sit at the top of the exposure list. About a third of the US workforce has near-zero exposure right now, mostly hands-on roles.

No mass layoffs since ChatGPT launched.
But hiring into high-exposure fields for 22-to-25-year-olds fell 14% in the same period. The youngest workers in those fields aren't being made redundant.
They're just not getting the entry-level roles that used to exist.

That's a slower, quieter form of disruption.

The opportunity side is worth sitting with.
The roles with the highest exposure (analysis, reporting, coding, data work) are exactly the ones where AI fluency turns into a compounding advantage. The person who can do the work and direct AI to do it faster is not at risk. They're the one getting hired.

The window to be that person is open right now. It won't stay open indefinitely!

Standard Claude is a conversation. You type, it responds, you implement.
CoWork is different.

You describe an outcome, point it at a folder on your computer, and it executes.
It reads your files, makes decisions, creates documents, organises data, and delivers finished output while you work on something else.

The simplest way to understand the gap:

Standard Claude says "here's how you'd prepare a variance report from these files."

CoWork says "I've read your P&L files, identified the top three variances, and drafted a two-page commentary ready for your board pack. I flagged one figure I wasn't sure about, can you confirm?"

That's not a subtle difference.

How to get it: CoWork lives inside the Claude Desktop app, not the browser. Go to claude.com/download. You need a paid plan. Pro at $20/month gets you access but you'll hit limits quickly on complex tasks.
For regular use, Max at $100/month is the realistic minimum.

First thing to do after installing: Point it at a folder. Click the folder icon, select a folder, and CoWork can now read and write everything inside it.
Nothing outside that folder.

Then go to Settings, find Global Instructions, and write a short brief: who you are, how you like to work, what language and style you prefer.
Claude applies that context to every session automatically.

Where it falls short (and you should know this): No memory between sessions, so save key context into a text file in your folder and ask it to read it first.
The desktop app must stay open while it works. And until now, no native M365 integration, though the Microsoft announcement above changes that picture for enterprise users.

My verdict: it is genuinely the most significant shift in how I work day to day.
The rough edges are real but they belong to a product that is three months old. The trajectory is clear.

Building is no longer the hard part. Any team with AI agents can ship.

Which means the thing that actually determines outcomes has shifted.

It's now judgment: what to build, in what order, and how to explain it clearly enough that people care.

The person who matters most in that environment isn't a manager or an engineer.

It's someone who can hold a clear picture of where the product should be in two years, work backwards from it, and tell the story in a way that organises the team and lands with users.

Technical enough to know what's possible.
Human enough to know what's worth building.

Almost no org is hiring like that's true. Most are still structured around engineering as the constraint. That constraint is gone.

The Anthropic jobs report makes the same point from a different angle.
The gap between what AI can do and what most orgs are actually doing with it exists largely because no one owns the vision and sequences toward it.
They have managers and executors. Not thinkers. You don't need more executors. You need a thinker at the top of the stack.