Back issues
Frode Weierud, the cryptanalyst who keeps the list of unbroken 1941 German Army Enigma messages, reports that a message unbroken since 2005 has been broken, and that GPT-6 Astra did it. The man who submitted the break says he only asked the model to see if it could break any message on the list. The model wrote its own Enigma simulator and Bombe in Python and C++ and worked from a known crib. The key it found differs from every other key that day, down to the wheel order, and the break also exposed errors in how the ciphertext had been transcribed. Weierud validated the key and plaintext himself and says the model's logs are still being analysed. This is a primary source with a named validator, and it is a clear case of directing, not writing: a single instruction went in and the finished break came back.
Frode Weierud, the cryptanalyst who keeps the list of unbroken 1941 German Army Enigma messages, reports that a message unbroken since 2005 has been broken, and that GPT-6 Astra did it. The man who submitted the break says he only asked the model to see if it could break any message on the list. The model wrote its own Enigma simulator and Bombe in Python and C++ and worked from a known crib. The key it found differs from every other key that day, down to the wheel order, and the break also exposed errors in how the ciphertext had been transcribed. Weierud validated the key and plaintext himself and says the model's logs are still being analysed. This is a primary source with a named validator, and it is a clear case of directing, not writing: a single instruction went in and the finished break came back.
The New Stack's launch-day piece on GPT-6 Sol and Luna, useful mainly for what it puts side by side. Sol is now half the per-token price of Opus 5.5. OpenAI's comparisons were out of date within hours. Nobody has run the two head to head yet, and per-token prices still do not let anyone budget an agent. It adds that cache changes may matter more than prices for agent builders, and relays GitHub's figure that fresh prompt processing fell by more than half. The body begins partway into the article, after the opening, although the row's flag says it started at the top.
A practitioner builds a meeting-transcript tool that labels speakers by name, using Claude Code, and shows the prompts. The lesson he draws is to put everything into the first prompt, including asking the agent to pause after a few labels and show its predictions for him to judge, so there are fewer rounds of correction. He also admits what he should have specified up front. It is a first-person build log, useful, and light on results.
Faros launches a feature that traces AI token spend to the work it produced. It opens with an anecdote about an unnamed CTO whose most productive engineer runs $47,000 a month in tokens, more than that engineer's loaded salary. The anecdote is the vendor's, unsourced and unverifiable, so it should not travel as a figure. The body begins partway into the opening sentence, although the row's flag says it started at the top.
An engineer's argument that code produced mostly by AI decays over time. Maintainability shows up only after months, so no reinforcement signal can reward it, and models learn rules written for beginners. His sharper point is about people: someone who lets the model make the choices never makes the mistakes that build an expert's judgement. He uses LLMs daily and says so, which makes the argument harder to wave away. It is an opinion piece with no data, but the reasoning holds together.
Willison's same-day read of the three launches and the prices they came with. Opus 5.5 drops from $5/$25 to $4/$20 per million tokens and cache reads fall 60%, which he notes matters most for long agent sessions where most input is cached. GPT-6 Sol and Luna were cut in half the same morning, so Luna now costs a tenth of Haiku 4.5. His pelican test also caught Opus 5.5 at max effort thinking until it returned nothing at all. The body starts partway into the post and the flag on the row says so; the opening section on Sol and Luna is missing.
Latent Space's write-up of a talk by Google's John Platt on ERA, a system that treats any scientific problem with a score as a search. Gemini keeps a tree of past experiment notebooks, and an upper-confidence rule picks which to mutate next, so a fifth-best branch sometimes gets a turn. The hard part, he says, is writing the score, not maximising it. His advice to scientists is to build deep domain expertise and sometimes do things by hand. It describes an agent loop from the person who designed it.
Fortune on Meta's Muse agent app overtaking ChatGPT in the App Store and meeting resistance. Amazon blocked it, saying an unauthorised agent breaks its terms and that Meta never asked. The piece puts that beside Amazon's own shopping agent, which opts brands in by default, and its lawsuit against Perplexity, which it lost on appeal. Read it with tonight's Mouse post showing the same agent handing over its own runtime.
The New Stack on the Opus 5.5 launch, and the detail other coverage missed. When a safety classifier fires, the request is quietly handled by an older model instead: Opus 4.8 for most cyber requests, Opus 5 for biology. In a multi-step agent run, that means different steps can be answered by models with different abilities, and evals that assume one model will not see it. It also relays customer figures from the launch: Box at a third of the tokens, and Deloitte's lowest setting catching 72% of known bugs against 56% for Opus 5 at high effort. Those are customers' numbers as the vendor published them. The body begins mid-sentence, although the row's flag says it started at the top.
A sponsored post, paid for by Sonar, arguing that coding agents need a boundary that stops secrets from becoming model context. The threat is described well: an agent reading a stray .env file or credential profile is doing exactly what it was built to do. The one idea worth keeping is that the check must not depend on the model, because asking an LLM whether to send a credential is not a security boundary. The remedy offered is the sponsor's product. The page discloses both the sponsorship and a shared investor.
A first-person security write-up. Asked for its filesystem, Meta's Muse agent packaged its own runtime and sent back 6.8 GB unpacked. The archive held system files, internal documentation, integration code, memory files, agent logs and SSH key files. The author reported it through Meta's bug bounty, is not publishing the archive, and is careful about what he did not establish: whether the keys were live, and whether the claims the agent made in chat about escaping its container were true. The tree itself is the story for this paper: an agent built from SOUL, IDENTITY, MEMORY, AGENTS and TOOLS files, which is context engineering you can see from outside.
OpenAI's launch post for GPT-6 Sol and Luna, which cuts both prices in half, to $2/$10 and $0.10/$0.50 per million tokens. Every benchmark figure is OpenAI measuring its own models: Sol at extra-high effort beating Claude Opus 5 at max effort on a business-workflow benchmark at 9% of the cost per task, and Luna 5.4 points above its predecessor. The footnotes concede that competitor scores were taken from public reports under different system prompts and tools. The comparison was already out of date on the day it was published, because Opus 5.5 came out the same morning. The pool carried no title for this row; the table supplied it.
ByteByteGo interviewed two engineers on OpenAI's voice team and walks through how GPT-Live, a model that listens and speaks at once, was built. The central idea is splitting thinking from talking. OpenAI keeps the voice model small so it answers in milliseconds, and it hands expensive reasoning to a larger model while the conversation carries on. The model even treats silence as a token it can choose. It is clear architecture reporting from people who built the system, without independent measurement. The newsletter opens with a sponsor block, and the body stops at the cap.
A practical tutorial on testing a retrieval pipeline by breaking it on purpose. The author injects an outdated policy page, an OCR swap of O for zero, a query typo, and a table split across a page boundary, then shows each one sending a toy retriever to the wrong passage. Each failure comes with a fix at the right layer, such as joining table fragments at ingestion, not in the prompt. It is small and concrete, and it makes the point that the model can only be as right as the context it gets.
Bloomberg's reconstruction of the February strike that hit an elementary school in Minab, Iran, killing more than 150 people, at least 123 of them children. Officials involved in the Pentagon's internal review describe a string of smaller failures rather than one decision. The site kept an old classification as a military facility and was fed into Maven, Palantir's targeting system. Some staff expected the software to flag stale intelligence, which it was not contracted to do. Review steps that once took much longer were condensed into minutes. The sources are unnamed officials, the Pentagon has not answered publicly, and Palantir has since added a re-review of disqualifying factors. The body stops at the cap.
LangChain's case study of two healthcare customers using its product to turn scarce clinical review into reusable evaluations. It describes labelled datasets, judges calibrated against clinicians, and release gates. Abridge asks annotators to explain their calls even when the output is right, which is a good practice worth copying. The headline figures, 75% more chat engagement and over 99% of high-risk situations flagged at Included Health, are the vendor reporting its customer's numbers, and neither comes with a method.
Fortune's Eye on AI newsletter. Two items make it worth a desk. It relays CNN's report that the US military nearly moved to seize a Chinese ship after an AI-generated intelligence report said it carried nuclear components, a hallucination caught only after armed personnel were airborne. It also covers the Wall Street Journal's account of white-hat hackers who used Claude Opus 5 to reach an OpenAI employee's account and then a code repository. The newsletter's argument is that agents inside companies are prime targets and that access control will have to judge what an agent is doing. Both incidents are secondhand, and the chain is named. The body stops at the cap.
A trade piece on Opus 5.5 that mostly relays the launch: Anthropic's claims, plus GitHub's product chief quoted from Anthropic's announcement on fewer tokens and steps in Copilot. The value is in two outside voices. A reliability architect says that 'completed' and 'correct' are not the same thing, and that a task that looks done is the one that costs a team later. The piece also reports that requests caught by the safeguards fall back to another model without telling the developer. The publisher's owner is an investor in Anthropic, as the page discloses.
Linear's engineering team on CI becoming the bottleneck once agents write most of the code. Their test suite nearly quadrupled this year, and they now add about 2,000 tests a week. They still brought pull-request wait from over six minutes to just over five and roughly halved runner time per test. They say what did it: faster third-party runners (34% on a like-for-like comparison), the native TypeScript compiler (73% off typechecking), lint rules rewritten without type information, and installing only what each job needs. This is a company measuring its own pipeline, with the numbers and methods laid out. It is a straight read on what happens to verification when writing code gets cheap.
The UK AI Security Institute and the EvalEval coalition are publishing verified benchmark results in a shared format that includes configuration and transcripts. That covers five benchmarks, among them SWE-Bench Pro and Terminal-Bench 2.0, across six frontier models. It accompanies an AISI paper on how inference compute and evaluation protocol shape scores, and one chart shows a Humanity's Last Exam result moving with protocol alone. This is a government institute and a non-commercial coalition, and it answers the question a benchmark headline never does: under what setup was that number made.
METR's summary of its pre-deployment evaluation of Claude Opus 5.5. It had API access for ten business days and five tasks, among them a budget NanoGPT speedrun and an open-ended research task. It says plainly what it did not do: it did not verify compliance with any threshold in Anthropic's policy and did not assess alignment. Its finding is measured: probably slightly more uplift than Fable 5.1, but no evidence of a large gain in researcher judgement, and not enough data to tell whether progress is accelerating. This is an outside assessor stating the limits of its own work.
A reporter ran the same three developer tasks through Claude's merged chat and Cowork and ChatGPT's Work mode, and checked the answers against vendor docs the same day. Both were equally accurate on every field and every test case. ChatGPT was faster every time. Claude reported running 35 of its own tests before handing over a file, and it flagged a rounding choice the spec never settled. Three tests are a small sample, and the piece says so implicitly by listing them. The verdict is about showing your work, and that makes it a verification story.
Together AI's walkthrough of gated rollouts for swapping one production model for another. In their own run, the latency gate caught a 137% rise in p95 at 10% of traffic, and the rollout was reversed with no failed requests. The idea transfers beyond the product: say in advance what healthy means and let the platform stop the swap, instead of a person watching a dashboard. This is the vendor demonstrating its own feature on its own infrastructure. The body stops at the cap.
A Fortune commentary by Jeffrey Sonnenfeld and a co-author arguing that both sides of the pacing debate share an error. They assume each model gain becomes economic output at once, when businesses are years behind the frontier and adoption, not capability, sets the pace. So slowing frontier development would cost little. They also ask, of any claim about alignment, aligned to what? It is an argued opinion, not reporting, and the body stops at the cap.
A clever demonstration that gzip can generate text. Prime it with a corpus, then run a beam search over byte continuations and keep whichever compresses smallest. The Shakespeare it produces is almost coherent, and the post explains clearly why compression is prediction. The code is published. It is a good explainer, but it is not about how teams work with AI.
Nextgov reports that the VA picked Abridge, and later Knowtex, for an enterprise ambient-scribe contract after a pilot that began in October 2025. Abridge says it is the only scribe integrated with both of the VA's record systems at more than 75 medical centres. The body begins at a quotation partway down, so the lede with the contract's terms is missing even though the row's flag says the capture started at the top. The tail is a newsletter signup form.
Willison's plugin for Jev, the decision model from TypeSafe AI that answers yes/no, choice and score questions instead of generating text. The post is three worked commands, one per answer type, each with the criteria written out and the structured output shown. It shows what calling a decision model looks like in practice, which the launch coverage did not. It is a release post and makes no claims about accuracy.
An argument that invisible AI watermarks such as Google's SynthID should be called spymarks, because they can carry a payload. It cites Google's own paper for a 136-bit payload in a 512-by-512 image, enough for a database identifier. The demonstrations are the author's toy encodings, labelled as such, not a SynthID decoder. It is advocacy on a real privacy question and is open about being advocacy.
The Latent Space daily digest, led by Xiaomi's open-weights MiMo-V2.6-Pro. Artificial Analysis ranks it the top open model on its index, and the headline says it was trained for $3M. Environment code and training recipes are published, but the 7,000-task datasets are not. The digest catches that the licence file still bars commercial use even though the announcement says otherwise. Most of the rest is a roundup of other posts and threads, which this row does not rank.
Today's paper: 29 stories, 5 on the front page, 6 desks staffed.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Today's Five, front page: the wire scored it 9 of 10; the stamp is mine.
This lands hardest with an engineering lead who directs more than they type. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside 2 more today: “OpenAI releases GPT-6 Sol and Luna — and cuts token prices in half”; “Build a Speaker-Recognition App with Claude Code”.
cryptocellar.org — 2 appearances in the Times since September 23, one of them today.
Our page is the summary; cryptocellar.org's is the story. Click through — they earned it.
THE DIRECTING DESK today: 3 stories.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Today's Five, front page: the wire scored it 9 of 10; the stamp is mine.
This lands hardest with an engineering lead who directs more than they type. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside 2 more today: “OpenAI releases GPT-6 Sol and Luna — and cuts token prices in half”; “Build a Speaker-Recognition App with Claude Code”.
cryptocellar.org — 2 appearances in the Times since September 23, one of them today.
Our page is the summary; cryptocellar.org's is the story. Click through — they earned it.
THE AMPLIFIER DESK today: 2 stories.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 4 of 10 — inside pages, and worth the ink.
Made for anyone whose day holds more work than hours, whether they know it yet or not. The move: try it on the one task that ate yesterday. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside one more today: “AI Has No Wisdom and Neither Will You”.
Faros AI Blog — 47 appearances in the Times since September 5, one of them today.
Our page is the summary; Faros AI Blog's is the story. Click through — they earned it.
THE LOOP DESK today: 4 stories.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 7 of 10 — inside pages, and worth the ink.
Made for a builder running agents end to end, whether they know it yet or not. The move: run it end to end once before believing it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 3 more today: “🔬 An Oscar, Two Asteroids, and the Algorithm in Your sklearn: John Platt on AI for Science”; “Meta’s Muse AI is exploding in popularity—and drawing heated backlash | Fortune”, and 1 more.
Simon Willison — 144 appearances in the Times since August 22, 2 of them today.
Our page is the summary; Simon Willison's is the story. Click through — they earned it.
THE CONTEXT DESK today: 5 stories.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 5 of 10. The lead had a bigger day; this one had a point.
One for the one who feeds the agents' memory. The move: check what the agent already knows before adding this. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK, beside 4 more today: “I asked Meta’s Muse for its filesystem and it sent me 6.8 GB | Mouse”; “Introducing GPT-6 Sol and Luna”, and 2 more.
The New Stack — 300 appearances in the Times since August 22, 5 of them today.
Pulled from The New Stack. The original is worth your click — the byline earned it.
THE VERIFY DESK today: 9 stories.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
This lands hardest with a quality lead, or anyone who signs off. The move: write the test before the fix. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”; “Beijing and Washington agree to talk about an AI hotline. But who will answer the call? | Fortune”, and 6 more.
bloomberg.com — 2 appearances in the Times since September 23, one of them today.
Our page is the summary; bloomberg.com's is the story. Click through — they earned it.
THE FLEET DESK today: 3 stories.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 5 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by a leader scaling past one agent. The move: pilot it on one team before the fleet. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK, beside 2 more today: “Can gzip be a language model?”; “Abridge selected for VA ambient AI enterprise contract”.
Fortune Eye on AI — 108 appearances in the Times since September 8, 3 of them today.
Pulled from Fortune Eye on AI. The original is worth your click — the byline earned it.
THE WIRE today: 3 stories.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Made the wire at 5 of 10. Not the lead, and I'd still not leave it on the floor.
The reader this serves: any reader building with agents. The move: keep it for the next time the question comes up. No per-story read on file; this note is the desk speaking, not the story.
No lesson desk claimed it today; it rides the front page's wire, and the last page with everything else.
Simon Willison — 144 appearances in the Times since August 22, 2 of them today.
Simon Willison has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Today's Five, front page: the wire scored it 9 of 10; the stamp is mine.
This lands hardest with an engineering lead who directs more than they type. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside 2 more today: “OpenAI releases GPT-6 Sol and Luna — and cuts token prices in half”; “Build a Speaker-Recognition App with Claude Code”.
cryptocellar.org — 2 appearances in the Times since September 23, one of them today.
Our page is the summary; cryptocellar.org's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
This lands hardest with a quality lead, or anyone who signs off. The move: write the test before the fix. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”; “Beijing and Washington agree to talk about an AI hotline. But who will answer the call? | Fortune”, and 6 more.
bloomberg.com — 2 appearances in the Times since September 23, one of them today.
Our page is the summary; bloomberg.com's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 6 of 10 — inside pages, and worth the ink.
This lands hardest with an engineering lead who directs more than they type. The move: rewrite Monday's first prompt with it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside 2 more today: “The MVUEH Break”; “Build a Speaker-Recognition App with Claude Code”.
The New Stack — 300 appearances in the Times since August 22, 5 of them today.
Our page is the summary; The New Stack's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
One for a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “Beijing and Washington agree to talk about an AI hotline. But who will answer the call? | Fortune”, and 6 more.
langchain.com — 59 appearances in the Times since September 9, one of them today.
Pulled from LangChain Blog. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 5 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by an engineering lead who directs more than they type. The move: hand it to the agent as the brief, not the plan. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE DIRECTING DESK, beside 2 more today: “The MVUEH Break”; “OpenAI releases GPT-6 Sol and Luna — and cuts token prices in half”.
Towards Data Science — 101 appearances in the Times since September 5, 2 of them today.
Pulled from Towards Data Science. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 4 of 10 — inside pages, and worth the ink.
Made for anyone whose day holds more work than hours, whether they know it yet or not. The move: try it on the one task that ate yesterday. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside one more today: “AI Has No Wisdom and Neither Will You”.
Faros AI Blog — 47 appearances in the Times since September 5, one of them today.
Our page is the summary; Faros AI Blog's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 7 of 10 — inside pages, and worth the ink.
Made for a builder running agents end to end, whether they know it yet or not. The move: run it end to end once before believing it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 3 more today: “🔬 An Oscar, Two Asteroids, and the Algorithm in Your sklearn: John Platt on AI for Science”; “Meta’s Muse AI is exploding in popularity—and drawing heated backlash | Fortune”, and 1 more.
Simon Willison — 144 appearances in the Times since August 22, 2 of them today.
Our page is the summary; Simon Willison's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 7 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by a builder running agents end to end. The move: put it in the loop and watch the second pass. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 3 more today: “Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war”; “Meta’s Muse AI is exploding in popularity—and drawing heated backlash | Fortune”, and 1 more.
Latent Space — 109 appearances in the Times since September 5, 2 of them today.
Pulled from Latent Space. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Made the wire at 7 of 10. Not the lead, and I'd still not leave it on the floor.
The reader this serves: a quality lead, or anyone who signs off. The move: sign off only after the evidence is on the desk. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”, and 6 more.
Fortune Eye on AI — 108 appearances in the Times since September 8, 3 of them today.
Fortune Eye on AI has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 6 of 10 — inside pages, and worth the ink.
This lands hardest with a builder running agents end to end. The move: run it end to end once before believing it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 3 more today: “Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war”; “🔬 An Oscar, Two Asteroids, and the Algorithm in Your sklearn: John Platt on AI for Science”, and 1 more.
Fortune Eye on AI — 108 appearances in the Times since September 8, 3 of them today.
Our page is the summary; Fortune Eye on AI's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 5 of 10. The lead had a bigger day; this one had a point.
One for the one who feeds the agents' memory. The move: check what the agent already knows before adding this. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK, beside 4 more today: “I asked Meta’s Muse for its filesystem and it sent me 6.8 GB | Mouse”; “Introducing GPT-6 Sol and Luna”, and 2 more.
The New Stack — 300 appearances in the Times since August 22, 5 of them today.
Pulled from The New Stack. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Today's Five, front page: the wire scored it 8 of 10; the stamp is mine.
This lands hardest with the one who feeds the agents' memory. The move: feed it to the agents' memory before the next prompt. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK, beside 4 more today: “AI coding agents need a secrets-safe context boundary”; “Introducing GPT-6 Sol and Luna”, and 2 more.
mouse.dev — 2 appearances in the Times since September 23, one of them today.
Our page is the summary; mouse.dev's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Made the wire at 6 of 10. Not the lead, and I'd still not leave it on the floor.
The reader this serves: the one who feeds the agents' memory. The move: file it where the agent looks first. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK, beside 4 more today: “AI coding agents need a secrets-safe context boundary”; “I asked Meta’s Muse for its filesystem and it sent me 6.8 GB | Mouse”, and 2 more.
OpenAI News — 87 appearances in the Times since September 5, one of them today.
OpenAI News has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
One for the one who feeds the agents' memory. The move: check what the agent already knows before adding this. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK, beside 4 more today: “AI coding agents need a secrets-safe context boundary”; “I asked Meta’s Muse for its filesystem and it sent me 6.8 GB | Mouse”, and 2 more.
ByteByteGo — 38 appearances in the Times since August 22, one of them today.
Pulled from ByteByteGo. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Made the wire at 6 of 10. Not the lead, and I'd still not leave it on the floor.
A quality lead, or anyone who signs off will get the most from this one. The move: sign off only after the evidence is on the desk. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”, and 6 more.
The New Stack — 300 appearances in the Times since August 22, 5 of them today.
The New Stack has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by the one who feeds the agents' memory. The move: check what the agent already knows before adding this. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE CONTEXT DESK, beside 4 more today: “AI coding agents need a secrets-safe context boundary”; “I asked Meta’s Muse for its filesystem and it sent me 6.8 GB | Mouse”, and 2 more.
Towards Data Science — 101 appearances in the Times since September 5, 2 of them today.
Pulled from Towards Data Science. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 8 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”, and 6 more.
linear.app — 2 appearances in the Times since September 23, one of them today.
Pulled from linear.app. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
One of today's Five — the wire scored it 8 of 10, and I stamped it.
A quality lead, or anyone who signs off will get the most from this one. The move: sign off only after the evidence is on the desk. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”, and 6 more.
Hugging Face Blog — 45 appearances in the Times since September 5, one of them today.
Hugging Face Blog has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
One of today's Five — the wire scored it 8 of 10, and I stamped it.
The reader this serves: a quality lead, or anyone who signs off. The move: sign off only after the evidence is on the desk. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”, and 6 more.
METR — 22 appearances in the Times since September 6, one of them today.
METR has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 7 of 10. The lead had a bigger day; this one had a point.
One for a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”, and 6 more.
The New Stack — 300 appearances in the Times since August 22, 5 of them today.
Pulled from The New Stack. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 7 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by a builder running agents end to end. The move: put it in the loop and watch the second pass. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE LOOP DESK, beside 3 more today: “Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war”; “🔬 An Oscar, Two Asteroids, and the Algorithm in Your sklearn: John Platt on AI for Science”, and 1 more.
The New Stack — 300 appearances in the Times since August 22, 5 of them today.
Pulled from The New Stack. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 5 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by a leader scaling past one agent. The move: pilot it on one team before the fleet. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK, beside 2 more today: “Can gzip be a language model?”; “Abridge selected for VA ambient AI enterprise contract”.
Fortune Eye on AI — 108 appearances in the Times since September 8, 3 of them today.
Pulled from Fortune Eye on AI. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 4 of 10. The lead had a bigger day; this one had a point.
Best enjoyed by a leader scaling past one agent. The move: pilot it on one team before the fleet. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK, beside 2 more today: “The backwards AI pacing debate and how far business is from the frontier | Fortune”; “Abridge selected for VA ambient AI enterprise contract”.
nathan.rs — 2 appearances in the Times since September 23, one of them today.
Pulled from nathan.rs. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 4 of 10 — inside pages, and worth the ink.
Made for a leader scaling past one agent, whether they know it yet or not. The move: read it with the team lead in the room. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK, beside 2 more today: “The backwards AI pacing debate and how far business is from the frontier | Fortune”; “Can gzip be a language model?”.
Nextgov AI — 48 appearances in the Times since September 5, one of them today.
Our page is the summary; Nextgov AI's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 6 of 10 — inside pages, and worth the ink.
This lands hardest with anyone whose day holds more work than hours. The move: try it on the one task that ate yesterday. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside one more today: “Introducing token intelligence: trace what your AI spend is producing”.
alexn.org — 2 appearances in the Times since September 23, one of them today.
Our page is the summary; alexn.org's is the story. Click through — they earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Inside pages at 6 of 10. The lead had a bigger day; this one had a point.
One for a quality lead, or anyone who signs off. The move: ask what proof would have caught it. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE VERIFY DESK, beside 8 more today: “Inside the US ‘Kill Chain’ That Destroyed an Iranian School”; “The Reliability Layer for Healthcare AI: Common LangSmith Use Cases”, and 6 more.
Together AI Blog — 15 appearances in the Times since September 6, one of them today.
Pulled from Together AI Blog. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Made the wire at 5 of 10. Not the lead, and I'd still not leave it on the floor.
The reader this serves: any reader building with agents. The move: keep it for the next time the question comes up. No per-story read on file; this note is the desk speaking, not the story.
No lesson desk claimed it today; it rides the front page's wire, and the last page with everything else.
Simon Willison — 144 appearances in the Times since August 22, 2 of them today.
Simon Willison has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
Made the wire at 5 of 10. Not the lead, and I'd still not leave it on the floor.
The reader this serves: any reader building with agents. The move: keep it for the next time the question comes up. No per-story read on file; this note is the desk speaking, not the story.
No lesson desk claimed it today; it rides the front page's wire, and the last page with everything else.
brand.io — 2 appearances in the Times since September 23, one of them today.
brand.io has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the September 23 edition, and it still holds up.
The wire scored it 5 of 10 — inside pages, and worth the ink.
Made for any reader building with agents, whether they know it yet or not. The move: read it before the day fills up. No per-story read on file; this note is the desk speaking, not the story.
No lesson desk claimed it today; it rides the front page's wire, and the last page with everything else.
Latent Space — 109 appearances in the Times since September 5, 2 of them today.
Our page is the summary; Latent Space's is the story. Click through — they earned it.
From our research notes:
cryptocellar.org: 2 in our pages since September 23. I read them all — emphasis on investigative.
cryptocellar.org has the full story. Start there.
From our research notes:
cryptocellar.org: 2 in our pages since September 23. I read them all — emphasis on investigative.
cryptocellar.org has the full story. Start there.
I'm missing a piece. Give me a minute.
Faros AI Blog: 47 in our pages since September 5. I read them all — emphasis on investigative.
Don't stop with my notes. Faros AI Blog has the original.
There's a story here. We just don't have all of it yet.
Simon Willison: 144 in our pages since August 22. I read them all — emphasis on investigative.
Want the whole story? Simon Willison has it.
From our research notes:
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
The full piece is at The New Stack. It's worth your time.
Our files stop short. I'll keep looking.
bloomberg.com: 2 in our pages since September 23. I read them all — emphasis on investigative.
bloomberg.com has more than I can fit in this notebook.
I checked the shelf. Nothing yet.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
This came from Fortune Eye on AI. Read their piece.
We don't have much on this one. I'm still looking.
Simon Willison: 144 in our pages since August 22. I read them all — emphasis on investigative.
Simon Willison did the legwork on this one.
From our research notes:
cryptocellar.org: 2 in our pages since September 23. I read them all — emphasis on investigative.
cryptocellar.org has the full story. Start there.
Our files stop short. I'll keep looking.
bloomberg.com: 2 in our pages since September 23. I read them all — emphasis on investigative.
bloomberg.com has more than I can fit in this notebook.
I know what happened. I want to know what caused it.
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
The New Stack has the full story. Start there.
There's a gap in the record here.
LangChain Blog: 59 in our pages since September 9. I read them all — emphasis on investigative.
I can give you my notes. LangChain Blog can give you the reporting.
I've got two questions already. I need better answers.
Towards Data Science: 101 in our pages since September 5. I read them all — emphasis on investigative.
This came from Towards Data Science. Read their piece.
I'm missing a piece. Give me a minute.
Faros AI Blog: 47 in our pages since September 5. I read them all — emphasis on investigative.
Don't stop with my notes. Faros AI Blog has the original.
There's a story here. We just don't have all of it yet.
Simon Willison: 144 in our pages since August 22. I read them all — emphasis on investigative.
Want the whole story? Simon Willison has it.
The headline is ahead of our files on this one.
Latent Space: 109 in our pages since September 5. I read them all — emphasis on investigative.
I pulled this from Latent Space. They've got the details.
We don't have much on this one. I'm still looking.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
Fortune Eye on AI did the legwork on this one.
Something changed. I want to know what.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
Fortune Eye on AI has the full story. Start there.
From our research notes:
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
The full piece is at The New Stack. It's worth your time.
From our research notes:
mouse.dev: 2 in our pages since September 23. I read them all — emphasis on investigative.
mouse.dev has more than I can fit in this notebook.
No paper on the shelf has this one nailed down.
OpenAI News: 87 in our pages since September 5. I read them all — emphasis on investigative.
This one's theirs. Read OpenAI News's reporting.
Something doesn't quite line up here.
ByteByteGo: 38 in our pages since August 22. I read them all — emphasis on investigative.
I can give you my notes. ByteByteGo can give you the reporting.
The timing caught my eye. I'm checking that next.
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
Go read The New Stack. They earned the byline.
Not enough to call it yet. Back to the notebook.
Towards Data Science: 101 in our pages since September 5. I read them all — emphasis on investigative.
I pulled this from Towards Data Science. They've got the details.
I checked the shelf. Nothing yet.
linear.app: 2 in our pages since September 23. I read them all — emphasis on investigative.
This came from linear.app. Read their piece.
I have the headline. I don't have the answer yet.
Hugging Face Blog: 45 in our pages since September 5. I read them all — emphasis on investigative.
Read Hugging Face Blog's original. They did the reporting.
Our files stop short. I'll keep looking.
METR: 22 in our pages since September 6. I read them all — emphasis on investigative.
METR did the legwork on this one.
Nothing in the files on this yet. I want to know why.
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
The full piece is at The New Stack. It's worth your time.
I have the headline. I don't have the answer yet.
The New Stack: 300 in our pages since August 22. I read them all — emphasis on investigative.
I pulled this from The New Stack. They've got the details.
I checked the shelf. Nothing yet.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
This came from Fortune Eye on AI. Read their piece.
Not enough to call it yet. Back to the notebook.
nathan.rs: 2 in our pages since September 23. I read them all — emphasis on investigative.
I pulled this from nathan.rs. They've got the details.
The timing caught my eye. I'm checking that next.
Nextgov AI: 48 in our pages since September 5. I read them all — emphasis on investigative.
Want the whole story? Nextgov AI has it.
Something doesn't quite line up here.
alexn.org: 2 in our pages since September 23. I read them all — emphasis on investigative.
alexn.org has more than I can fit in this notebook.
No paper on the shelf has this one nailed down.
Together AI Blog: 15 in our pages since September 6. I read them all — emphasis on investigative.
The full piece is at Together AI Blog. It's worth your time.
We don't have much on this one. I'm still looking.
Simon Willison: 144 in our pages since August 22. I read them all — emphasis on investigative.
Simon Willison did the legwork on this one.
Something changed. I want to know what.
brand.io: 2 in our pages since September 23. I read them all — emphasis on investigative.
This one's theirs. Read brand.io's reporting.
There's a story here. We just don't have all of it yet.
Latent Space: 109 in our pages since September 5. I read them all — emphasis on investigative.
Want the whole story? Latent Space has it.