Key Takeaways
- Ramp data from 70,000 businesses: Anthropic's flagship Fable 5 accounts for roughly 11% of spend — cheaper models released in July and a full generation older have already passed it
- Pew: 34% of US adults now use AI chatbots for at least one health-related task; 72% say it's extremely or very important that a provider discloses when AI is used in their care
- Salesforce announced its CRM now runs inside Claude directly — sales teams can pull records, review deals, and update information without opening Salesforce itself
- A federal judge ruled the Pentagon unlawfully retaliated against Anthropic for refusing to override its limits on lethal weapons and mass surveillance
- Everything this week points the same direction: AI is getting less exotic — cheaper than the flagship, embedded rather than separate, operating under rules that hold up
Week of August 24–30, 2026
Four things happened in AI between August 24th and the 30th that are worth five minutes of your time as someone who runs a business.
None of them ask anything of you. But together they say something fairly specific about where this is settling.
This is Last Week in AI — the signal, not the noise. Four stories operators need to understand, and what each one means for how you run your business.
Story 1: The best AI on the market is being ignored by the companies paying for it
The payments company Ramp tracks what 70,000 businesses actually spend money on. The Financial Times got hold of their numbers on AI, and they're not what anyone expected.
Anthropic's Fable 5 is the most advanced model the company sells. It launched in June to a lot of noise. More than two months later, it accounts for roughly 11% of what businesses spend on Anthropic's products.
The cheaper option they released in late July has already passed it. So has a model that's a full generation older — that one pulls in three and a half times as much money as the flagship.
The reason isn't performance. Fable 5 costs about double its nearest competitor, and businesses kept finding the cheaper thing did the job.
A partner at Accel, an investment firm that put close to a billion dollars into Anthropic, put it plainly to the FT: "Most people don't need to operate at the frontier." The period where customers automatically reached for the most advanced option, he said, "was not a durable era."
What it means for operators: if you've been putting this off because you assumed doing it properly meant buying the serious version, 70,000 companies just tested that assumption with their own money and concluded otherwise. (Same lesson as which AI model your business should use: pick what fits the job, not the leaderboard.)
Story 2: A third of Americans are now asking AI about their health
Pew Research published a survey of about 3,500 American adults. 34% now use AI chatbots for at least one health-related task — looking something up, trying to work out what's causing a symptom, getting a plain-English explanation of a lab result.
Nearly half say the answers are very helpful. Only 29% are comfortable sharing personal health information with these tools.
The number worth writing down: 72% of Americans say it's extremely or very important that a doctor or healthcare provider tells them if AI is being used in their care.
Pew published a second report the same day with a harder finding. On people using chatbots for loneliness, depression, or stress, Americans are roughly twice as likely to say the tools hurt rather than help.
What it means for operators: your customers are already using AI to research things before they ever contact you. In healthcare that's explicit, but the same behaviour applies to anyone comparing services or working out whether they need a professional at all.
And that 72% is the clearest read yet on how people want this handled. Not banned. Disclosed. The public isn't asking businesses to avoid AI. They're asking to be told when it's in the room. That aligns with what we wrote last week about keeping AI in the back office — and it's a much easier standard to meet than abstinence.
Story 3: Salesforce moved its entire CRM inside Claude
Salesforce, the biggest customer-management software company in the world, announced its system now runs inside Claude directly. Sales teams can pull up customer records, review deals, prepare for meetings, and update information without ever opening Salesforce itself.
The commitment behind it isn't small. Salesforce plans to spend roughly $300 million with Anthropic this year on top of an existing stake. Their share price jumped double digits.
If you don't use Salesforce, the product doesn't matter to you. The direction does.
What it means for operators: for three years the model has been that AI lives somewhere separate. You open a chatbot, you type, you copy the answer back into whatever you were actually doing.
That's now reversing. AI is being built into the software businesses already run, so the tool comes to you rather than you going to the tool. A fair amount of the "getting started" problem is about to solve itself — the gap between using AI and running it is narrowing inside the tools you already pay for.
Story 4: A judge told the Pentagon it can't punish an AI company for having rules
This one got less attention than it deserved.
Anthropic refused to sign an open-ended Pentagon contract that would have overridden the limits it places on how Claude can be used. The company holds two lines publicly: it won't knowingly support autonomous lethal weapons, and it won't support domestic mass surveillance.
The Defense Department responded by labelling Anthropic a "supply chain risk" — a designation normally reserved for foreign adversaries — which would have blocked federal agencies and contractors from using its technology at all.
On Friday a federal judge ruled that was unlawful retaliation. The Pentagon may appeal, and NPR reports it could eventually reach the Supreme Court.
What it means for operators: when you pick an AI vendor you're also picking their rules about what the tool can and can't be used for. Until this week it wasn't clear those rules meant much under real pressure. A court has now said a company can hold its own limits without being punished out of the market for it.
That matters more than it sounds. The terms your provider sets around your data and how it gets used are only worth something if they can survive somebody powerful leaning on them.
Also this week
Nvidia reported another record quarter on Thursday, then quietly paused some of the financing deals that had been keeping smaller AI companies afloat. Worth noting mainly as a signal: the money behind this is getting more selective about who it backs.
Putting it together
The most powerful AI is losing to the cheaper one. The public wants disclosure rather than abstinence. The technology is moving inside the tools people already use. And a court just backed a vendor's right to hold its own limits.
Everything this week points the same direction: AI is getting less exotic. Cheaper than the flagship, embedded rather than separate, and starting to operate under rules that hold up.
If you've been waiting for this to become normal before you engage with it, that's roughly what normal looks like. It doesn't arrive with an announcement. It's the expensive option quietly losing, and the useful version quietly turning up inside the software you already pay for.
— Aristotle Taylor, CEO & Co-Founder, Levron Labs
What to do this week
If you're paying for a premium AI tier, ask whether the cheaper option in the same product family would cover what you actually do with it. Most owners never check — and 70,000 businesses just showed why that's worth five minutes.
Not sure where AI belongs in your stack versus where a human should stay in the loop? Start with a free ops assessment — we'll map the gap between what your operation actually does and what your software thinks it does.
Sources
Businesses choosing cheaper models over the flagship (Aug 24)
Pew Research on health chatbot use (Aug 25)
Salesforce and Anthropic (Aug 26)
Pentagon ruling (Aug 28)