3 weeks ago the US government gave a tech company 90 minutes to switch off its own product, and on Wednesday it let them switch it back on.

I've spent 17 years in corporate finance and I've sat through some strange announcements in that time, but I'd never watched a government treat a piece of software like this before, so I went and read everything Anthropic published about it, partly out of curiosity and partly because the thing they switched back on is now sitting in my Claude account with a deadline attached.

Here's the story as plainly as I can tell it. Anthropic is the company behind Claude, and on 9 June they released a new engine for it called Fable 5, the most capable one they've ever put out. 3 days later the US Commerce Department ordered it offline over security concerns, and after nearly 3 weeks of negotiating it came back this Wednesday. If you pay for Claude there's nothing to install, it's already sitting in your model menu, and until Monday 7 July it's included in your plan for up to half your weekly usage. After Monday you pay for it separately, per use. One thing worth checking if your company runs Claude through an Enterprise account, because standard seats don't get that included week at all, so it's worth asking whoever manages your account whether it's switched on before you rely on it.

It was the per use part I wanted to understand properly before writing any of this, because it's measured in something called tokens, and I wanted to be able to explain that without it turning into a computer science lesson. A token is about three quarters of a word, so a million tokens is roughly 1,500 pages, and you pay one rate for everything the model reads and 5 times that rate for everything it writes back. The catch with Fable 5 is that it also thinks before it answers, privately, sometimes for minutes, and every word of that hidden thinking is billed at the writing rate, which is why it drains your allowance faster than the other models even when the answer on your screen is only 3 lines long.

That looks steep until you put it against a real piece of work, so here are my rough sums for a few actual jobs on Fable 5, with the hidden thinking counted in:

Look at where the money actually sits, because the biggest job on that list costs less than a sandwich. The place the cost quietly runs away from you isn't the big jobs, it's habit, leaving Fable 5 as your default and letting it think hard about a question that Haiku could have answered in a second. Use it like a specialist you bring in for the heavy stuff and the numbers stay almost boring.

So here's the one job I'd give it this week while it still costs nothing. Dig out the last 10 things you wrote to win something, proposals and quotes if you run a business, business cases and budget requests if you're employed, and include the ones that lost as well as the ones that won, because the losers are where the useful information is hiding. Upload the whole lot in one message, tell it which ones won, and ask what the winners have in common, what the losers have in common, and what you do in your own writing that you're too close to notice. Then set it one rule, that every pattern it claims it has to back up by quoting your own sentences to you, because that rule is the difference between real analysis and a machine quietly agreeing with you.

I've just launched something and wanted to share with you all.

It's called the AI Audit. If you read this issue and thought, fine, I believe it works, but where does it actually start in my work... that is the exact question it answers.

Here's what it is. You and me on a Zoom call for 45 to 60 minutes, talking through how you actually work. No preparation needed. Within 48 hours you get a 7 to 9 page report built around your specific role or business, with four prioritised recommendations, the implementation detail for each, and the ROI worked out... the hours and the money, per week. Then a 30 minute follow-up call where we walk through it together.

One version for professionals, one for business owners.

For professionals: configurai.com/professionals-audit.html
For business owners: configurai.com/business-owners-audit.html

See you next week with more ways to use AI without losing your mind (or your credibility).

Orgesa Meli

P.S. If this saved you from a future hallucination disaster, forward it to someone who's using ChatGPT/Claude for proposals, reports, or client work. They'll thank you later. Subscribe to my community here.