Read time: 5 minutes

Leader, welcome back.

Sunday edition this week instead of the usual Wed/Thu, the CFO news was too good to sit on.

Anthropic's own CFO just told a podcast that Claude writes his statutory financials. A human checks the output. Claude does the work. That's the story this week, everything else is the supporting cast.

Krishna Rao, Anthropic's CFO, went on Invest Like the Best and described how his finance team actually uses Claude day to day.

Statutory financial statements for every one of Anthropic's legal entities are now produced by Claude, with a human checking the output rather than building it.

His team has built a library of more than 70 finance-specific skills.
On top of that sits one skill called MFR, the monthly financial review, which Rao says gets him 90 to 95% of the way there.

His words: it turns the discussion from "what happened" into "what do we do about it." A weekly revenue report that used to take hours now takes 30 minutes.
And the heaviest user on his team isn't the 22-year-old with a coding background, it's the head of tax.

Here's why I'm leading with this over any product launch.
This is not a vendor claiming AI will transform finance.
This is the CFO of the company that makes the model, describing what his own team does with it, on the record.
If you've been sitting on the fence about whether this is real, that's a hard thing to argue with.

The part I'd push back on before you get excited: 90 to 95% ready is doing a lot of work. The 5 to 10% a human still checks is exactly the part that matters most, the judgment calls, the exceptions, the things that don't fit the pattern.

Rao isn't claiming Claude replaced his team's thinking.
He's claiming it removed the mechanical part so the thinking has somewhere to go. That's the whole point of everything I teach in the course, and it's genuinely useful to have Anthropic's own CFO making the case for me.

In brief, the rest of the week:

Claude Cowork is merging into the main Claude.ai chat.
No more separate tab for bigger jobs, your connected apps and context now live in one place, and it can keep working after you close your laptop.
Expect OpenAI to copy this within a couple of months.

Claude Artifacts is spinning out dedicated Docs and Slides products, with Claude Design moving into regular conversations too.
Put next to the Cowork merge, the pattern is obvious: Anthropic is building a workspace, not a chatbot. Office and G Suite should be paying attention.

Claude Code's newest feature has a name now, Claude Mods. They change how Claude Code looks and behaves, and you can ask Claude to write the mod for you. Minor, but a nice example of AI configuring the tool you use to run AI.

CFOs are quietly tightening the reins on AI agents with access to ERPs, billing systems and bank portals. The pattern showing up across finance leaders: inventory every agent, set expiry dates so nothing runs unreviewed, and require human sign-off on anything irreversible like a new beneficiary or a payment change.
If you've given any AI tool standing access to a financial system, this is worth a five-minute audit this week.

I'm testing Jev this week, the new model from TypeSafe AI built by one of ChatGPT's co-creators.

Instead of generating text, it spits out probabilities across a fixed set of answers, which makes it fast and cheap for the kind of repeated yes/no or multiple-choice judgment calls that would be overkill for a full LLM call.

If interested in deeper dive, reply and let me know.

Signal: Gemini API shipped two new Live models, 3.8 Live and 3.8 Live Extended Thinking, taking video and audio in and returning audio, built for real-time assistance. I ran a few minutes in AI Studio without a single hiccup. At 7x cheaper than GPT-Live 1, worth a test if you need a model that talks back live.

Noise: A think piece doing the rounds arguing it's time to retire double-entry bookkeeping. Read as what it is, a well-produced advertorial for a spend management platform, not a serious proposal. Skip it.