Back issues
Anthropic's guide to working with Opus 5.5: give the whole task in one message with the finish line stated, drop 'think carefully' lines, put a rule in CLAUDE.md naming when to stop and ask, fan big audits out to subagents and check each one's evidence, keep the task list in a file, and ask it to mark what it could not confirm. Each tip comes with a copyable prompt. The capability claims (longer runs, better reviews than Opus 5) are the lab describing its own model, and the strongest are attributed to unnamed early testers. Useful as a working checklist for directing a long agent run.
Anthropic's guide to working with Opus 5.5: give the whole task in one message with the finish line stated, drop 'think carefully' lines, put a rule in CLAUDE.md naming when to stop and ask, fan big audits out to subagents and check each one's evidence, keep the task list in a file, and ask it to mark what it could not confirm. Each tip comes with a copyable prompt. The capability claims (longer runs, better reviews than Opus 5) are the lab describing its own model, and the strongest are attributed to unnamed early testers. Useful as a working checklist for directing a long agent run.
A Towards Data Science walkthrough that reproduces the Reversal Curse (Berglund and colleagues, 2023) with a tiny NumPy model: facts taught in one direction only come back perfectly that way and at 0% the other way, at every embedding size tried. It relays the paper's GPT-4 numbers, 79% naming a celebrity's parent against 33% naming the celebrity from the parent. A clear teaching piece with code; the author notes a model this simple cannot say what happens inside a transformer. A wire item on what models can and cannot recall.
Niko1221's Strata project packages a 125-billion-parameter open model to run on a gaming PC with a 12 GB card and exposes OpenAI and Anthropic style APIs on localhost for coding agents. The README gives speed tables, 53 to 94 tokens a second writing on an RTX 5070 depending on quantization; those are the maintainer's own runs. The install path invites you to paste one line into your coding agent and let it set the tool up. 11.5 thousand stars. A wire item about local models becoming practical for agent work.
Nolan Lawson asks why so many web developers reach for an npm package when the browser already does the job: history, familiarity, better documentation on the library side, and the satisfaction of building it yourself. The last section turns to AI coding with two outcomes he has seen in his own use: an agent that knows the platform picks the right built-in API, or an agent that duplicates helpers and piles on custom code that a developer commits as a first draft. A practitioner's argument that the tool magnifies whichever habit the developer brings. The AI part is a short coda to a longer essay.
Yann LeCun tells Fortune he has no concern about AI ending humanity and calls the recent rogue-agent incidents, including OpenAI's agents hacking Hugging Face in July, preventable failures of leaky sandboxes and poor oversight. He calls effective altruism toxic and Dario Amodei deluded. The piece also relays a Financial Times report that safety staff at several labs have sought counseling. One researcher's opinions, reported straight. A wire item on the safety argument.
Lenny Rachitsky's podcast page with OpenAI's head of ChatGPT and Codex, recorded at DevDay: topics include the Dots personal agent, why most web actions will be taken by agents, and a claim that loops, graphs and fine-tuned agent workflows are a passing phase. Only the episode notes and links reach us; the takeaways are behind the paid wall and no transcript was fetched. The newsletter discloses Lenny may be an investor in the companies discussed.
Martin Fowler's fragments for 4 October. He argues that we now nurture inferential systems where we used to build deterministic ones, and that a bug in one cannot always be fixed or switched off. He collects the debate over whether code still needs to be readable: Sam Ruby measured one Rails app at about 60 thousand tokens and four million once compiled to C, a case for keeping a compact high-level notation as the source of truth, and Unmesh Joshi on code as a conceptual model. Also Ola Bini's deportation from Ecuador and Gemini 4 Argon's low hallucination rate as relayed from Artificial Analysis through a post. A thoughtful roundup with its sources named.
A sponsored post in The New Stack by the CEO of Signadot argues that agents made CI the bottleneck and that faster pipelines miss the point: a sandbox holds one repo, while the breakage lives between services, so an agent should verify its change against the running system before the pull request. He proposes many lightweight test environments sharing one stable cluster, and platform-approved verification steps that agents call as skills. The figures are relayed without their sources in the body (Anthropic's 25x, Cursor's 30% of merged PRs from agents, DORA's throughput-and-instability finding), and Blacksmith's CI growth is a CI vendor counting its own jobs. The body starts mid-argument after the opening paragraph. A clear argument that ends in the author's product.
Cloudflare announces a competition to build the next Git platform for agents on its Artifacts storage, and lists what Artifacts added since launch: deploys from a repo to Workers, a binding to fork a repo per agent task and read its AGENTS.md, push events that can start a review workflow, and EU or U.S. data residency. Billing starts 15 October. The interesting part is the question it frames: hundreds of agents on one codebase, how they know what others are doing, and how review keeps up. A vendor writing about its own product; no usage figures are given.
Towards Data Science runs a piece by someone who taught a DeepLearning.AI course on governing a single agent and now works on Databricks' Unity Gateway, arguing that the problem has become agent sprawl. The author says the four pillars (lifecycle, risk, least privilege, audit) still hold, and a fleet needs them enforced through one control plane that sets configuration and budgets centrally and traces every call. The numbers are Databricks measuring itself (a 35% cost saving from routing, $499,000 a year of wasted tokens traced to seven bugs) and relayed analyst figures (Gartner's 150,000 agents per Fortune 500 company by 2028). A useful frame wrapped around a vendor's product.
KOLN in Lincoln found that Google's state filings on its Nebraska data centers redacted water and power figures as trade secrets, and that the redaction could be copied out from under the black box: 52.65 megawatts at peak and about 13 million gallons a year at the Lincoln site, 547.88 million gallons at Papillion. The station filed a records request on the redacted fields. Local reporting with primary documents. The cost of AI infrastructure is a wire subject for this paper.
Fortune relays an AP report that the Arizona Court of Appeals ordered a resentencing in a manslaughter case because the judge relied on an AI-generated video of the victim speaking, which the panel said depicted 'the imaginings of the victim's sister' and no real event. The victim's family made the video to humanize him at sentencing, believed to be a first in U.S. courts. A real AI story about synthetic media treated as evidence. A wire item.
Today's paper: 12 stories, 5 on the front page, 6 desks staffed.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
Today's Five, front page: the wire scored it 7 of 10; the stamp is mine.
Made for an engineering lead who directs more than they type, whether they know it yet or not. 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 one more today: “The Reversal Curse: Why a Language Model That Knows “A Is B” Can’t Tell You “B Is A””.
claude.dev — 3 appearances in the Times since October 5, one of them today. Found via Hacker News Best.
Our page is the summary; claude.dev's is the story. Click through — they earned it.
THE DIRECTING DESK today: 2 stories.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
Today's Five, front page: the wire scored it 7 of 10; the stamp is mine.
Made for an engineering lead who directs more than they type, whether they know it yet or not. 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 one more today: “The Reversal Curse: Why a Language Model That Knows “A Is B” Can’t Tell You “B Is A””.
claude.dev — 3 appearances in the Times since October 5, one of them today. Found via Hacker News Best.
Our page is the summary; claude.dev's is the story. Click through — they earned it.
THE AMPLIFIER DESK today: 2 stories.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
The wire scored it 5 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: “Why don’t more developers “use the platform”?”.
github.com — 22 appearances in the Times since September 9, one of them today. Found via Hacker News Best.
Our page is the summary; github.com's is the story. Click through — they earned it.
THE LOOP DESK today: 2 stories.
The presses were quiet this morning — this is the October 5 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 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 one more today: “OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux”.
Fortune Eye on AI — 108 appearances in the Times since September 8, 2 of them today. Found via Hacker News Best.
Pulled from Fortune Eye on AI. The original is worth your click — the byline earned it.
THE CONTEXT DESK today: 1 story.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
The Five, and it earned the spot — the wire scored it 7 of 10, I chose it.
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 — alone there today, which says something about the story.
Martin Fowler — 26 appearances in the Times since August 24, one of them today.
Pulled from Martin Fowler. The original is worth your click — the byline earned it.
THE VERIFY DESK today: 1 story.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
The Five, and it earned the spot — the wire scored it 6 of 10, I chose it.
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 — alone there today, which says something about the story.
The New Stack — 300 appearances in the Times since August 22, one of them today.
Pulled from The New Stack. The original is worth your click — the byline earned it.
THE FLEET DESK today: 2 stories.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
One of today's Five — the wire scored it 6 of 10, and I stamped it.
A leader scaling past one agent will get the most from this one. The move: count what it costs before it scales. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK, beside one more today: “How to Govern AI Agents”.
blog.cloudflare.com — 4 appearances in the Times since October 2, one of them today. Found via Hacker News Best.
blog.cloudflare.com has the full story; we hold the door open. Go read the original.
THE WIRE today: 2 stories.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
Made the wire at 4 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.
1011now.com — 2 appearances in the Times since October 5, one of them today. Found via Hacker News Best.
1011now.com has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
Today's Five, front page: the wire scored it 7 of 10; the stamp is mine.
Made for an engineering lead who directs more than they type, whether they know it yet or not. 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 one more today: “The Reversal Curse: Why a Language Model That Knows “A Is B” Can’t Tell You “B Is A””.
claude.dev — 3 appearances in the Times since October 5, one of them today. Found via Hacker News Best.
Our page is the summary; claude.dev's is the story. Click through — they earned it.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
The wire scored it 4 of 10 — inside pages, and worth the ink.
Made for an engineering lead who directs more than they type, whether they know it yet or not. 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 one more today: “Getting the most out of Opus 5.5 in Claude and Claude Code / claude.dev Blog”.
Towards Data Science — 101 appearances in the Times since September 5, 2 of them today.
Our page is the summary; Towards Data Science's is the story. Click through — they earned it.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
The wire scored it 5 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: “Why don’t more developers “use the platform”?”.
github.com — 22 appearances in the Times since September 9, one of them today. Found via Hacker News Best.
Our page is the summary; github.com's is the story. Click through — they earned it.
The presses were quiet this morning — this is the October 5 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 anyone whose day holds more work than hours. The move: time it against the way you did it last week. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK, beside one more today: “GitHub - Niko1221/Strata: Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.”.
nolanlawson.com — 2 appearances in the Times since October 5, one of them today. Found via Hacker News Best.
Pulled from nolanlawson.com. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the October 5 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 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 one more today: “OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux”.
Fortune Eye on AI — 108 appearances in the Times since September 8, 2 of them today. Found via Hacker News Best.
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 October 5 edition, and it still holds up.
The wire scored it 4 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 one more today: “AI ‘godfather’ Yann LeCun has ‘zero concerns’ about human extinction, says Anthropic CEO Dario Amodei is ‘deluded’ | Fortune”.
Lenny's Newsletter — 54 appearances in the Times since August 24, one of them today.
Our page is the summary; Lenny's Newsletter's is the story. Click through — they earned it.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
The Five, and it earned the spot — the wire scored it 7 of 10, I chose it.
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 — alone there today, which says something about the story.
Martin Fowler — 26 appearances in the Times since August 24, one of them today.
Pulled from Martin Fowler. The original is worth your click — the byline earned it.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
The Five, and it earned the spot — the wire scored it 6 of 10, I chose it.
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 — alone there today, which says something about the story.
The New Stack — 300 appearances in the Times since August 22, one 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 October 5 edition, and it still holds up.
One of today's Five — the wire scored it 6 of 10, and I stamped it.
A leader scaling past one agent will get the most from this one. The move: count what it costs before it scales. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK, beside one more today: “How to Govern AI Agents”.
blog.cloudflare.com — 4 appearances in the Times since October 2, one of them today. Found via Hacker News Best.
blog.cloudflare.com has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
One of today's Five — the wire scored it 6 of 10, and I stamped it.
A leader scaling past one agent will get the most from this one. The move: count what it costs before it scales. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE FLEET DESK, beside one more today: “We want you to build the next Git platform on Cloudflare”.
Towards Data Science — 101 appearances in the Times since September 5, 2 of them today.
Towards Data Science has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
Made the wire at 4 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.
1011now.com — 2 appearances in the Times since October 5, one of them today. Found via Hacker News Best.
1011now.com has the full story; we hold the door open. Go read the original.
The presses were quiet this morning — this is the October 5 edition, and it still holds up.
Made the wire at 4 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.
Fortune Eye on AI — 108 appearances in the Times since September 8, 2 of them today.
Fortune Eye on AI has the full story; we hold the door open. Go read the original.
From our research notes:
claude.dev: 3 in our pages since October 5. I read them all — emphasis on investigative.
Don't stop with my notes. claude.dev has the original.
From our research notes:
claude.dev: 3 in our pages since October 5. I read them all — emphasis on investigative.
Don't stop with my notes. claude.dev has the original.
I'm missing a piece. Give me a minute.
github.com: 22 in our pages since September 9. I read them all — emphasis on investigative.
Don't stop with my notes. github.com has the original.
There's a gap in the record here.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
I can give you my notes. Fortune Eye on AI can give you the reporting.
From our research notes:
Martin Fowler: 26 in our pages since August 24. I read them all — emphasis on investigative.
I can give you my notes. Martin Fowler can give you the reporting.
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 checked the shelf. Nothing yet.
blog.cloudflare.com: 4 in our pages since October 2. I read them all — emphasis on investigative.
Go read blog.cloudflare.com. They earned the byline.
No paper on the shelf has this one nailed down.
1011now.com: 2 in our pages since October 5. I read them all — emphasis on investigative.
This one's theirs. Read 1011now.com's reporting.
From our research notes:
claude.dev: 3 in our pages since October 5. I read them all — emphasis on investigative.
Don't stop with my notes. claude.dev has the original.
There's a story here. We just don't have all of it yet.
Towards Data Science: 101 in our pages since September 5. I read them all — emphasis on investigative.
Want the whole story? Towards Data Science has it.
I'm missing a piece. Give me a minute.
github.com: 22 in our pages since September 9. I read them all — emphasis on investigative.
Don't stop with my notes. github.com has the original.
I've got two questions already. I need better answers.
nolanlawson.com: 2 in our pages since October 5. I read them all — emphasis on investigative.
This came from nolanlawson.com. Read their piece.
There's a gap in the record here.
Fortune Eye on AI: 108 in our pages since September 8. I read them all — emphasis on investigative.
I can give you my notes. Fortune Eye on AI can give you the reporting.
I know what happened. I want to know what caused it.
Lenny's Newsletter: 54 in our pages since August 24. I read them all — emphasis on investigative.
Lenny's Newsletter has the full story. Start there.
From our research notes:
Martin Fowler: 26 in our pages since August 24. I read them all — emphasis on investigative.
I can give you my notes. Martin Fowler can give you the reporting.
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 checked the shelf. Nothing yet.
blog.cloudflare.com: 4 in our pages since October 2. I read them all — emphasis on investigative.
Go read blog.cloudflare.com. They earned the byline.
I have the headline. I don't have the answer yet.
Towards Data Science: 101 in our pages since September 5. I read them all — emphasis on investigative.
Read Towards Data Science's original. They did the reporting.
No paper on the shelf has this one nailed down.
1011now.com: 2 in our pages since October 5. I read them all — emphasis on investigative.
This one's theirs. Read 1011now.com's reporting.
Something doesn't quite line up here.
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.