Read time: 4 minutes

Leader, welcome back!

Most people use AI like a vending machine.

Put in a question, get out an answer, move on.

That works - if all you want is "good enough."
But the gap between "good enough" and "genuinely useful" is almost always in how you ask.

I teach this in my AI for Finance Leaders course, and the moment it clicks for people is always the same: when they see the difference between a bad prompt and a good one, side by side, on a real task.

So let me show you.

Let's say you need to analyse a P&L before a meeting tomorrow.

❌ What most people type:

You'll get a generic summary.
Technically correct, completely useless in a boardroom!
It'll tell you revenue went up and costs went down - things you could see by looking at the spreadsheet for 10 seconds.

✅ What you should type instead:

The difference in output is night and day.

The first prompt gives you a book report.
The second gives you something you'd actually send to your CFO.

Why it works:
This follows what I call the RACEF framework - Role, Action, Context, Examples, Format.
Five elements that turn any AI prompt from vague to precise.

The one most people skip is the E - giving AI an example of the tone, structure, or style you want. It's often the difference between output you'd rewrite from scratch and output you'd send with minor edits.

That's not a hack.
That's how the best AI users work every single time.
Once you see it, you can't unsee it.

This is the core framework from Module 1 of my course.
The full version goes deeper - with a 64-prompt library covering everything from variance commentary to board pack narratives, advanced techniques and complete workflows for forecasting and decision memos.

It's currently available at a discounted rate of £99 for a limited period of time!
Check it out here.

Now - the gap between "good enough" and "genuinely useful" isn't just about prompting. It's about speed.

I read something this week from Claire Vo that nailed it: businesses that look like they're adopting AI are often just decorating the surface while everything underneath stays slow.

The real test isn't "do you use AI?"
It's "can your team do things same-day that used to take a week?"

Fix a bug.
Launch a landing page.
Turn around a negotiation.
Upgrade to a new model.

Before AI, slowness was understandable.
Now? If you can move fast and you don't - that's not a resource problem.
That's a culture problem.

Keep that lens on as you read this week's stories.

Anthropic dropped Sonnet 4.6 on Monday.
Second major release in two weeks.

The headline: it approaches Opus-level intelligence at one-fifth the price.
Complex document analysis, multi-step financial work, navigating spreadsheets - all now running at the cheaper tier. One enterprise tester saw reasoning accuracy jump from 62% to 77%.

My take: AI is getting smarter and cheaper simultaneously.
Every month, the "good enough" tier improves. If you've been waiting for the right moment - the excuses are thinning out!

I'm running Claude Max personally, which gives me extended access to both Opus and Sonnet plus Claude Code for building directly from my terminal.
Paired with Cursor (via my GPT Pro sub) for a second angle on coding tasks. Between the two, I can prototype and ship things in a weekend that would have taken a team weeks. More on my exact setup in a future issue.

OpenClaw is the open-source project that lets you build a personal AI agent - one that browses the web, manages files, and operates your apps autonomously. Built by one person, used by thousands.

This weekend, OpenAI announced they'd brought the creator on board.
The project moves to a foundation with OpenAI's backing.

Why this matters: OpenClaw was impressive but too technical for most people - you needed a separate machine, terminal knowledge, and comfort managing security yourself. Now it's getting the resources and distribution of the biggest AI company in the world.

The direction is clear: AI agents that don't just answer questions but do things for you are moving from demo to product, fast.
If you're on Claude, you can already feel this with CoWork - AI that acts, not just advises. If you would like a deeper dive on this, please do let me know.

I haven't used OpenClaw yet but will be tinkering with it. Watch this space.

Signal: → Anthropic raised $30B at a $380B valuation - more than double six months ago. The AI arms race is accelerating, not slowing.

→ Deutsche Bank asked AI how it plans to disrupt jobs. Answer: wealth management first, with robo-advisors potentially advising 80% of retail investors by 2027. If you're in finance, this isn't distant.

Noise: → Another cycle of "AI will take all jobs" panic.
The nuance nobody writes: AI is changing how work gets done, not eliminating it. The people who learn to work with it will do more, earn more, and stress less.
That's the whole point of this newsletter!