Read time: 4 minutes

Leader, leader, welcome back!!

Some news first, then the stuff worth bookmarking.

Four things from the past fortnight that should be on your radar.

1. Anthropic just teamed up with Goldman, Blackstone, and Apollo.
$1.5B venture. The plan: embed Anthropic engineers inside mid-sized businesses to actually deploy Claude into real workflows.
The headline isn't the money. It's the signal.
The bottleneck has officially moved from the models to the deployment.
Big money is now paying for the exact operationalisation problem most teams are still pretending isn't theirs.

2. 76% of organisations now have a Chief AI Officer.
Up from 26% last year, per IBM's 2026 survey. The role isn't about building models. It's about getting the org to actually adopt them. If you don't have someone owning this, three out of four of your competitors do. Worth a conversation with your leadership team this month.

3. Claude paid plans just got double the usage.
Anthropic doubled the limits on Pro, Max, Team, and Enterprise plans, and scrapped the slowdown that used to hit during peak hours. If you bounced off the cap before, worth another look.

4. Gemini now gives you the actual file.
Ask it for a doc, sheet, slide deck, or PDF and it hands you the download instead of chat text. Small upgrade, real productivity win for anyone living in Google Workspace.

You don't need to write a line of code to use this.

But if you're trying to figure out how to actually deploy AI on your team (what to staff, how to structure the work, how to delegate to AI without ending up with one bloated chatbot nobody can explain), this is the cheat code.

I've watched finance teams burn six months building a single mega-prompt before someone works this out. Don't be that team.

Any AI system you hand to a team should have three layers, not one.

Layer 1: Skills.
The atomic units of work. One skill, one job.
"Draft a variance commentary."
"Triage overdue invoices."
"Summarise this week's board pack inputs."
If you can't describe what it does in seven words without using "and then," it's actually two skills. Build and test each one on its own.

Layer 2: Commands.
Recipes that chain skills together.
/run-month-end might trigger the variance skill, then the commentary draft skill, then the board pack summary, with you approving at every handoff.
Don't chain more than three skills per command.
Past four, the agent starts drifting.

(How to actually build a Skill: next section.)

Layer 3: The orchestrator.
This is the layer most teams skip.
It listens to natural language ("I'm stressed about the board pack on Friday"), matches it to the right command, explains what it's about to do, and runs on confirmation. The orchestrator never does the work itself. It only routes.
The moment it starts pulling data or drafting commentary on its own, you're back to a bloated agent that everyone is afraid to touch.

The teams whose AI work scales can answer three questions.
What atomic jobs do our agents do?
Who owns each one?
What's the routing layer that picks between them?

The teams whose AI work plateaus built one mega-prompt that handles "everything." Three months later nobody can explain what it does.

Same pattern applies whether you're building an internal finance agent or evaluating a vendor pitch. Next time someone sells you an "AI agent," ask what their orchestrator routes between and what each underlying skill does. If they look confused, that's your answer.

Monday morning move. Get your team to list the 10 most common requests their workflow handles today. Each is a candidate skill. Group them into 3 to 5 workflows. Each workflow is a candidate command. Then figure out who owns the routing layer. That's your starting framework.

Claude has a feature called Skills. Reusable instruction sets you install once and trigger with a slash command, in any chat. Think of it as saving a prompt permanently to Claude's brain.

If you have prompts you run weekly (variance commentary, board pack drafts, invoice chasing emails), this kills the copy-paste tax forever.

How to set one up.
In Claude, click Customize, then the + button, then Create Skill.
Structure it in three parts.
A description and when Claude should trigger it.
Your instructions. Two or three examples of good output.
Save it, then type /[skill name] in any chat to activate.

Rule for when to bother (credit Alex Pereira). The task has to be recurring, structured, and have a consistent output format. All three? Build the skill.
One or two? Just save it as a prompt.

Starter template:

For when a week has happened, a lot has moved, and you need someone to think with you. Paste this into Claude with your raw weekly notes:

The "uncomfortable truth" line is the part that earns its keep. Try it once.