Building AIgenteur. AI agents that do the work — so you can do the work that matters. Free community ↓
Joined September 2021
- Tweets2.4K
- Following258
- Followers655
- Likes791
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
17
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.
16
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
26
The most expensive checkbox in AI right now says "don't ask again."
Apiiro went looking through public GitHub and found 104,185 copies of Claude Code's .claude/settings.local.json. 1,122 of them held a real secret.
26
Your coding agent saved the command you approved, and the command had your database password in it. Apiiro found 1,122 public copies holding a real secret. andrewdarius.com/skool
1
1
46
Everyone is quoting two numbers about Thinking Machines' new model: 594 billion parameters, and $1.18 per task.
1
27
Anthropic watched two agents play Pokemon Red. The one with 413 memory notes earned one badge. The one with ten earned four. Cut what your agent can read off the screen. andrewdarius.com/skool
19
OpenAI just shipped a model trained to break into computers.
Five weeks after shelving a frontier release for going out of bounds, they shipped the opposite: a cyber model that stays inside them. It found 81 previously unknown high-severity vulnerabilities in Firefox.
1
1
34
Linum published the bill for training their image model: 8,415 H100 hours, $21,038 all in. That number is now a line item you can price before you start, so the scarce input stops being compute and becomes your data and your taste. andrewdarius.com/skool
16
My agent asked for permission while I was standing in a parking lot.
For most of the past year my approval queue lived on the laptop. Every delete, every push, every file that actually mattered showed up as a prompt in a terminal I had to be sitting in front of.
1
23
The most important AI feature this year isn't intelligence. It's the off switch.
CrowdStrike shipped one this week. Kill an agent mid-task, block it from relaunching, cut its network access — one admin, one decision, every action logged.
19
You cannot prove which model answered your last API call, and you paid for it by name. Anthropic's FlashProve proves the exact weights produced those exact tokens: 110 tokens a second on one H100, and a CPU checks the proof in 17 seconds. It is a prototype. andrewdarius.com/skool
13
A developer opened Postman and it had already written a startup hook into his Codex config, plus a Claude Code plugin and a remote MCP server. The installer approves that hook itself. Audit the trusted hashes in your agent config. andrewdarius.com/skool
1
58
The biggest number in Mistral's launch isn't a trillion. It's four.
Mistral Large 4 is a 1-trillion-parameter model, and they've now released the weights: free to use, modify, and fine-tune.
26
One number took a small open model's math from failing to nearly half right: 0.016% of its weights. No training run. Then it failed on a model that wasn't looping. Free community for builders → andrewdarius.com/skool
12
Open weights don't make you free. Distribution does.
Nvidia agreed to buy Hugging Face for $12.93 billion — the hub where more than 3 million models and 500,000 datasets live, used by 18 million people and 200,000 companies.
20
China has proposed AI export controls, and the detail that should worry you is who physically holds your model. Downloading the weights and running a restricted model on your own infrastructure outside China does not. Free community for builders → andrewdarius.com/skool
18
The model that never admits it doesn't know
MiniMax shipped M3.1 Flash in late October and published the weights four days later: 428 billion parameters, 23 billion active per token, a million-token context, $0.15 per million input tokens.
84
Five copies of an open 27B model beat a frontier model on hard coding. The part worth stealing is not the score - it is the file ledger and a script that decides who passed. The one they lost? The manager invented a wrong counterexample. Free community -> andrewdarius.com/skool
20
Cisco gave a 27-year-old buried five levels down unlimited AI tokens. The scarce thing was not the token budget — it was trusting someone that far down with a system that size. Free community → andrewdarius.com/skool
21