Three things happened this week that were not supposed to happen yet. An AI started drafting hospital reports. The bills for autonomous agents came due. And the men who predicted the AI jobs apocalypse quietly took it back. Follow the thread; it ends somewhere interesting.
The AI that waited eight years
Aidoc’s model has watched brain scans in 2,000 hospitals since 2018. Flagging emergencies, saying nothing. This week it picked up a pen: it now writes the first draft of the report your doctor signs. Read what changed →
It is not working alone. On a live NHS ward, another AI is predicting infections before symptoms show. See the pilot →
The $1.3 million experiment
Machines that work send bills. One developer let 100 coding agents run free for a month, and they spent 603 billion tokens of his money. The invoice: $1.3 million. Where it went should worry anyone with an AI budget. See the breakdown →
$26 billion for a coding agent
Investors read the same invoices and bid higher. Devin’s valuation doubled to $26 billion in ten months, and underneath sits a revenue multiple nobody wants to say out loud. Do the math →
Then the price war started
Numbers like that invite attackers. Cursor now claims frontier coding at a tenth of the price, and the same week OpenAI made Codex pay-per-token with no way out for new business seats. Two moves, one squeeze on your budget. Compare the costs →
“Pretty wrong”
And while the machines were getting hired and priced, the men who promised they would take your job blinked. “Pretty wrong” is Sam Altman’s own verdict on his predictions; Dario Amodei softened his too. The question is not what they said. It is why both said it now. Read the timing →
Second Opinion
The physician’s read on the week: the risk in AI-drafted radiology reports is not the writing, it is the anchoring. A doctor editing a machine’s draft reads differently from one starting blank, and the liability still lands on the human signature. Watch who audits the drafts, not who writes them. The full story →
One number
67 percent of clinicians already use AI every day. The debate about whether this should happen is still running. Reality left early. See how deep it goes →
One thing to use
Kimi. Remember those cheap coding models above? There is a decent chance the one inside your tool is Kimi, a Chinese trillion-parameter family almost nobody checks for. Meet the model →
That is the thread: the machines got jobs, the jobs got bills, and the prophets got humble. Hit reply and tell me which story deserves the full treatment next; I read everything. Forwarded this? Claim your own copy.
Dr. Joseph Joshua
P.S. Next issue: the flaw that scored a perfect 10.
AI and business tech news, verified by a physician who reads the filings. One email a week, no noise.
