Treat Intelligence as Borrowed
The Departure
Last week, I told you to build your summer road-warrior stack on Claude Fable 5. Three days later, the government pulled it.
That's not an exaggeration. Fable 5 launched June 9. On June 12, the Commerce Department issued an export-control order, and rather than build nationality-based access controls, Anthropic withdrew the model entirely. The "free through June 22" window I sent you? Gone before it ever closed.
So let me own that, and then let me make it useful. Your meeting-brief workflow from Dispatch #022 still works. The pipeline ranking from Dispatch #020, the close plans from Dispatch #021: all still good. The workflows were never the problem. The problem was wiring them to one model that could vanish with 72 hours notice. This week, we fix that for good.
The Co-Pilot
Tool: Claude Opus 4.8 via claude.ai. Stable, available, unaffected by any of last week's mess. If you ran the #022 brief on Fable 5, move it back here, and it works again at no change in cost.
The Use Case: A paste-and-go audit that tells you which of your AI workflows would break tomorrow if the tool behind them disappeared, and how to rebuild each one so it can't.
The Prompt (paste two things first: the AI workflows you actually rely on, and the tool or model each one runs on. Then add this):
"You are an operations advisor. Here are the AI workflows I rely on and the tool each one runs on: [paste]. For each workflow: (1) flag whether it would break tomorrow if that single tool or model disappeared; (2) tell me whether the real value is a durable artifact I keep (a brief, template, or plan) or a live dependency that only works while the tool is up; (3) name a fallback model or tool I could switch to within an hour; (4) rewrite the workflow so its output survives any one tool being pulled. Output as a table: workflow | single point of failure? | durable or live | fallback | resilient version."
New here, or this got forwarded to you? You don't need any prior dispatch. List the three AI things you do most in a week, name the tool for each, and run it cold. The audit works without any setup.
And it runs on your phone. Type your workflows into the Claude app from a gate or a back seat, run it, and save the table to Notes. That table is the point.
The Upgrade
Topic: Why durable beats live, plus two things worth a look
The readers who came out of the Fable 5 pull just fine were the ones who saved the brief. Not the ones who depended on the model. As one developer put it this week, treat intelligence as borrowed: use the best model while you have it to produce a durable artifact you keep, then run it later on whatever's cheap and available. A brief is yours. A plan is yours. A live dependency on one model is a loan that can get called. Build for the thing you walk away with.
Microsoft Copilot Cowork went generally available on June 17. If your team pays for Microsoft 365, you got an AI agent layer yesterday at no extra subscription cost. Microsoft claims it's 30-40% cheaper per prompt than Claude alternatives. That's a vendor's number, not a tested one, so don't move your stack on it. Spend 30 minutes to see whether it includes a tool you already pay for separately before you buy another standalone AI sub.
One more reason not to bet everything on a single tool. SemiAnalysis data out this month shows that a $200/month Claude Max plan can cost Anthropic up to $8,000/month in compute. Labs are subsidizing heavy users 10 to 40 times over. As they march toward IPOs, those prices rise, not fall. Lock nothing in.
Skipping this week: the SpaceX/Cursor $60B deal and Z AI's GLM-5.2 open weights. Both are interesting, but neither one changes what you do before your next flight.
The Landing
Under 20 minutes, today. Take your Dispatch #022 meeting-brief workflow, or whatever your single most important AI workflow is, and do two things. Move it to Opus 4.8 so it runs again. Then save its output as a file you keep: a brief in Notes, a doc in your drive, a printed page in your bag.
That's the whole principle in one task. The model is borrowed. The artifact is yours.
Q3 budgets open in about two weeks. Walk in with a stack you've stress-tested instead of one you're hoping holds, and the next model that gets pulled is somebody else's bad morning, not yours.
Safe travels,
Marcellus