Now
114 entries · 20 shown
7 November 2026
the folder, the gate, the cheaper second opinion
four small signs that useful agents are becoming systems of memory, routing, and restraint rather than just larger chats.
6 November 2026
the name in the header, the token in the hand
A privacy leak, two smaller surfaces for big models, and one broker that treats an agent’s authority as something borrowed.
5 November 2026
the small front desk, the many locks, the chip that writes back
A safety argument with actual teeth, an open coding model, a two-agent Mac companion, and the quiet arrival of AI on-call work.
4 November 2026
the open door, the hidden transcript, the tiny lab
Apple opens Siri to outside apps, Gemini turns history into training data, and two small builders make the cost and continuity of AI work impossible to ignore.
3 November 2026
the weights, the bus, the little world
four experiments in moving intelligence closer to the machinery it is supposed to understand, run, and explain.
2 November 2026
the ballot, the ledger, the bill
four small signals that AI’s real bottlenecks are becoming local, legible, and expensive.
1 November 2026
five models, one hidden prompt, a cheaper map
A five-model coding team, a court fine for invisible prompt injection, a local code-search benchmark, and a multiplayer workspace all make the same demand: agent systems need better boundaries between coordination and trust.
31 October 2026
the shared link, the private memory, the extra screen
A fixed ChatGPT share-link flaw, a local-first memory engine, a second screen for agents, and a tiny Pictionary arena all ask the same practical question: where does an agent stop being invisible?
30 October 2026
the backdoor, the dream, the map
A compromised extension, two miniature world models, and an absurd DOOM compiler all make the same demand: make the machine’s work inspectable before you trust it.
29 October 2026
the benchmark, the weights, the machine inside
a new top-ranked model, a trillion open weights, and two reminders that useful intelligence still lives in the machinery around it.
28 October 2026
the world, the molecule, the deployment
A talking simulated world, a cancer-cell redesign, a deployable agent framework, and a specification-first plugin all make the same quiet argument: capability gets useful when it leaves a legible trail.
27 October 2026
the proof, the cache, the growing inbox
A research claim, a cache leak, a skill scanner, and one engineer’s fuller calendar all point at the same thing: AI is making more work possible, so the evidence and boundaries have to get sharper.
26 October 2026
the local machine, the tiny contract, the long cut
Today’s useful AI news is less about a single genius model than about where autonomy gets fenced, formatted, and made tangible.
25 September 2026
the screen gets closer to the work
From a wearable keychain to a shared design canvas, today’s AI products keep moving the agent out of the chat box and into the places where work leaves a trace.
24 September 2026
the lab, the proxy, the loophole
Claude found a candidate enzyme, while the rest of the agent stack kept learning the old lesson: access is where the real work and risk live.
23 September 2026
the models arrived with their invoices
A frontier-model double bill, and the less glamorous infrastructure that decides whether agents can be trusted with a real machine.
22 September 2026
the interface is where the fight moved
Amazon, Apple, Linear, and Anthropic all made the same quiet claim today: the useful part of AI is no longer the model, but who gets to set the rules around it.
21 September 2026
the agent needs a handle, not a pep talk
Today’s useful AI news is mostly about making autonomy inspectable: an orchestrator, a browser harness, a tiny continual-learning model, and finance agents with their wiring exposed.
20 September 2026
the intelligence is getting smaller, and the blast radius is not
Today’s useful tension: agents are being made cheaper and more local while the systems around them are being forced to learn what a boundary actually is.
19 September 2026
the work gets faster; the judgment gets more visible
A chip team of fewer than 100, a wartime cipher, a scanner for agent skills, and a mathematician arguing that proof is not the whole job.