Microsoft's 365 AI agent ships from a team of 11 developers.
At the DealBook Summit this week, Nadella was asked about the 60,000 buyouts.
6
KPMG ran the same audit twice: once with its accountants, once with more than 100 AI agents. Its UK head of audit says the AI went deeper and asked better questions. KPMG says the record goes down to individual characters and timestamps.
Free community → andrewdarius.com/skool
34
Asked for a painting as close to a photo as possible, Claude Fable 5.2 wrote a 1,000 to 2,000 line Python program that places every brushstroke itself, roughly 700,000 of them for one parking-garage scene.
The program is the deliverable.
Free community → andrewdarius.com/skool
29
Every AI agent failure I've dealt with wasn't a model failure. It was an evidence failure.
Microsoft just launched an AI Transparency Initiative at Ignite with more than 40 organizations behind it, and one line in that announcement is the sharpest thing written about…
1
56
A TV channel just aired a full sales show with nobody in the room.
Hyundai Home Shopping in Korea put its new system, AI-REAL, on air this week. It generates the entire program — voice and visuals — with no host standing in front of a lens and no camera shoot.
25
The line you still type into prompts is now dead weight.
Anthropic says delete "think hard": thinking is always on in Opus 5.5, and removing those lines made replies start sooner with no clear drop in quality.
Free community → andrewdarius.com/skool
22
Alibaba just shipped its flagship closed. Moonshot shipped its weights.
Qwen4-Max is the strongest model Alibaba has ever built: a 2.4-trillion-parameter backbone with 71 billion active per token, a million-token context window, native vision.
1
25
My agent can't buy anything. That's not a capability problem.
Every time I hand an agent something real — reserve a room, buy a licence, settle a charge — it dies at the same checkpoint. Not because the model can't do the work.
19
Claude now sits in your video call as a visible participant, and the catch is the one thing it does not save.
It keeps a transcript, a summary and screen descriptions.
Free community → andrewdarius.com/skool
29
My agents have never failed at reasoning. They keep failing at doors.
United Airlines just announced that third-party AI agents aren't welcome on its site — citing shared account credentials, unintended purchases, and FAA/DOT rules that require a human to tick the…
1
1
63
AWS shipped a developer tool where its own team wrote and reviewed zero lines of code.
Three agent teams, tests checked against real AWS, a merge gate on every change.
Swami spent four nights supervising an agent that could not check its results.
andrewdarius.com/skool
22
The first AI invoice I ever got, I couldn't explain.
The tokens were cheap. The bill wasn't. Every single turn, my agent re-sent the whole conversation — instructions, tool definitions, the same forty pages of context — and I paid for all of it again.
1
1
22
48 agents went through 1,200 square kilometers of lidar in one night and flagged a hilltop with straight edges.
Then the update landed. Archaeologists had already mapped that hilltop in a 2025 paper, and his agents had missed it in the literature.
andrewdarius.com/skool
30
The most expensive word in automation is "I can't help with that."
Reflection published the full safety numbers behind Beam, their 501B open-weight model.
16
Your business can now claim a page inside ChatGPT, and the catch is one rule about what you sell.
Claim the profile in Business Hub, put real prices in the product feed, then read the discovery topics — they show how AI already talks about your business.
andrewdarius.com/skool
17
Both labs said the quiet part out loud this week.
Meta shipped Muse Ember — its smallest, cheapest model — and described it as built for "the work agents repeat all day." Tool calls. Routing. Extraction. Parsing long documents.
1
1
34
Tokens per ticket fell from six thousand seven hundred and twenty-four to two thousand and ten, and about two point two cents a ticket became about a twentieth of a cent. andrewdarius.com/skool
28
8% vs 77%. Same AI agents. The difference isn't the model.
Zapier just published its 2026 AI Agents survey — 500+ enterprise leaders. 71% now have agents in production, up from 40% a year ago.
24
Claude Code and GitHub now go to Goldman's engineers — more than twelve thousand of roughly forty-seven thousand staff. Traders and bankers lost them. andrewdarius.com/skool
27
