Back issues
Ollama's post launching Phonon text-to-speech models on Apple Silicon, with English and four other languages, an OpenAI-compatible speech endpoint, and streaming playback piped from a chat model. Ollama measures itself at 35% faster than Gradium's own runtime on an M5 Max and reports word error rate matching the reference implementation; both are the house counting its own shelves. The 40M models are noncommercial. A local-model launch for the wire; the capture opens partway in.
Quiet on this desk today — nothing on the wire fit it. The front page holds the day.
Ollama's post launching Phonon text-to-speech models on Apple Silicon, with English and four other languages, an OpenAI-compatible speech endpoint, and streaming playback piped from a chat model. Ollama measures itself at 35% faster than Gradium's own runtime on an M5 Max and reports word error rate matching the reference implementation; both are the house counting its own shelves. The 40M models are noncommercial. A local-model launch for the wire; the capture opens partway in.
Quiet on this desk today — nothing on the wire fit it. The front page holds the day.
A Towards Data Science guide to six architecture patterns for getting enterprise data ready for agents, written from practice but without code or results. The strong parts: re-embedding driven by change-data-capture with tombstones so deleted documents leave the vector store, carrying source access controls into retrieval as filters applied before search, and a certified metric layer so the model does not invent joins or definitions. It ends on an agent refusing to query an uncertified table, framed as a feature. The body stops at the capture cap.
A Towards Data Science practitioner piece on the agent that keeps answering after its tool starts returning 401s, ending with a confident answer and a success status. The fix is in code, outside the prompt: sort errors into retryable and not, cap retries with jitter, hand the model a one-line TOOL_FAILED instruction, and refuse at the end of the run if no tool call succeeded. The author names the guard's own hole (one success in five passes it) and a bug in their first version that treated an empty result as failure. Their dashboard advice: count the refusals, and if the count sits at zero for months, check the wiring. The Gartner cancellation figure is relayed.
The New Stack on the PHP2Shell emergency patches, with the disclosure fight on the record: Chainguard shipped patched images six hours before the embargo after rediscovering the flaw with the prompt "find bad vulnerabilities in PHP", and the PHP Foundation called that irresponsible. Maintainers from Atuin, Valkey and OpenSSF say the hard part has moved from finding bugs to fixing them, and that small projects with unpaid maintainers cannot absorb the flood. The 493-issues figure is Anthropic's own research about its own model, relayed here. The outlet discloses that its owner invests in OpenAI, Wiz and Anthropic.
A New Stack head-to-head of DeepSeek V4.1-Flash-1017 and Claude Haiku 5.5 on three graded tests: an agentic bug fix, 20 SQL questions with traps, and triage of 400 error reports. DeepSeek scored 6 of 6, 20 of 20 and 1,600 of 1,600 against Haiku's 5, 20 and 1,596, at $0.07 total against $0.30. Most of the cost gap came from one long prompt that crossed Haiku's 100,000-token price step, and Haiku's prompt caching was left off. The writer is a developer marketing writer at InfluxData, independent of both vendors, and ran each test once, which the piece itself says proves little. The capture opens partway in, past the lede.
LangChain on the internal agents it runs across sales, engineering and marketing, and the five things each needed: living in Slack, acting on behalf of the user for some tools and under a fixed identity for others, global and per-user memory, code run in a sandbox behind a secret proxy, and mostly scheduled or event-triggered runs. The specifics are the value, such as email tools available only in DMs and a code reviewer that posts under its own GitHub app. A vendor describing its own use of its own product, which it pitches at the end. A primary account of a working setup, with no metrics.
A New Stack post written by Ory, which sells agent identity, arguing that agent security needs authorization at the point of action, not just valid credentials. The useful example: an agent deleting a production database with a developer's real credentials passes network, runtime and identity checks while doing something the human never meant. It proposes a cycle of observe, enforce, build and deploy, with monitoring first and deterministic policy around probabilistic behavior, placed in the harness. The argument is sound and the vendor sells the answer. The body stops at the capture cap.
Latent Space's write-up of an Anthropic session on dynamic workflows in Claude Managed Agents, where the lead agent plans the work and writes an orchestration program that launches sub-agents. The Q&A is the useful part: the model does not know its budget while running but developers can set one, sub-agents share a filesystem but cannot message each other, flows cost much more than one agent, and a dreaming feature reorganizes memory between runs. Every claim comes from Anthropic's presenters about their own product. Dated October 12, a late arrival, and the capture opens partway in.
A CNCF-sponsored New Stack post by Dan Ciruli of Nutanix on running agents on Kubernetes. It argues for sandboxing agents beyond the container, explicit permission grants for files, tools and network, and a warm pool of suspended pods so short-lived agents start in milliseconds. It relays a CNCF survey figure (two-thirds of organizations running inference do it on Kubernetes) and refers to agents escaping lab test environments without naming a source. It closes by pointing readers to commercial distributions, which is what the author's employer sells. The capture opens mid-sentence.
Today's paper: 9 stories, 5 on the front page, 4 desks staffed.
One of today's Five — the wire scored it 4 of 10, and I stamped it.
Anyone whose day holds more work than hours will get the most from this one. The move: pass it to whoever is out of hours. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK — alone there today, which says something about the story.
Ollama Blog — 16 appearances in the Times since September 9, one of them today.
Ollama Blog has the full story; we hold the door open. Go read the original.
THE DIRECTING DESK today: quiet — nothing on the wire fit it.
Nothing landed here today. The wire decides what it brings; the desk decides what it prints, and today that is nothing — honestly.
THE AMPLIFIER DESK today: 1 story.
One of today's Five — the wire scored it 4 of 10, and I stamped it.
Anyone whose day holds more work than hours will get the most from this one. The move: pass it to whoever is out of hours. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK — alone there today, which says something about the story.
Ollama Blog — 16 appearances in the Times since September 9, one of them today.
Ollama Blog has the full story; we hold the door open. Go read the original.
THE LOOP DESK today: quiet — nothing on the wire fit it.
Nothing landed here today. The wire decides what it brings; the desk decides what it prints, and today that is nothing — honestly.
THE CONTEXT DESK today: 1 story.
Inside pages at 5 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 — alone there today, which says something about the story.
Towards Data Science — 105 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 VERIFY DESK today: 3 stories.
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 2 more today: “PHP’s security scramble is just a preview: AI is digging up flaws faster than maintainers can fix them”; “DeepSeek V4.1-Flash-1017 vs. Claude Haiku 5.5: More accurate at a fraction of the price”.
Towards Data Science — 105 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 FLEET DESK today: 4 stories.
Today's Five, front page: the wire scored it 7 of 10; the stamp is mine.
This lands hardest with a leader scaling past one agent. 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 3 more today: “AI agent security is a lifecycle — not a gatekeeper”; “How Anthropic’s Dynamic Workflows Coordinate Claude Agents (Updated with video!) — Latent Space”, and 1 more.
langchain.com — 61 appearances in the Times since September 9, one of them today.
Our page is the summary; LangChain Blog's is the story. Click through — they earned it.
One of today's Five — the wire scored it 4 of 10, and I stamped it.
Anyone whose day holds more work than hours will get the most from this one. The move: pass it to whoever is out of hours. No per-story read on file; this note is the desk speaking, not the story.
Filed on THE AMPLIFIER DESK — alone there today, which says something about the story.
Ollama Blog — 16 appearances in the Times since September 9, one of them today.
Ollama Blog has the full story; we hold the door open. Go read the original.
Inside pages at 5 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 — alone there today, which says something about the story.
Towards Data Science — 105 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.
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 2 more today: “PHP’s security scramble is just a preview: AI is digging up flaws faster than maintainers can fix them”; “DeepSeek V4.1-Flash-1017 vs. Claude Haiku 5.5: More accurate at a fraction of the price”.
Towards Data Science — 105 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.
One of today's Five — the wire scored it 7 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 2 more today: “Agentic AI in Production: Why Your Agent Keeps Answering After Its Tools Fail”; “DeepSeek V4.1-Flash-1017 vs. Claude Haiku 5.5: More accurate at a fraction of the price”.
The New Stack — 318 appearances in the Times since August 22, 4 of them today.
The New Stack has the full story; we hold the door open. Go read the original.
Made the wire at 6 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 2 more today: “Agentic AI in Production: Why Your Agent Keeps Answering After Its Tools Fail”; “PHP’s security scramble is just a preview: AI is digging up flaws faster than maintainers can fix them”.
The New Stack — 318 appearances in the Times since August 22, 4 of them today.
The New Stack has the full story; we hold the door open. Go read the original.
Today's Five, front page: the wire scored it 7 of 10; the stamp is mine.
This lands hardest with a leader scaling past one agent. 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 3 more today: “AI agent security is a lifecycle — not a gatekeeper”; “How Anthropic’s Dynamic Workflows Coordinate Claude Agents (Updated with video!) — Latent Space”, and 1 more.
langchain.com — 61 appearances in the Times since September 9, one of them today.
Our page is the summary; LangChain Blog's is the story. Click through — they earned it.
The wire scored it 6 of 10 — inside pages, and worth the ink.
This lands hardest with a leader scaling past one agent. 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 3 more today: “The five components of building a "company harness"”; “How Anthropic’s Dynamic Workflows Coordinate Claude Agents (Updated with video!) — Latent Space”, and 1 more.
The New Stack — 318 appearances in the Times since August 22, 4 of them today.
Our page is the summary; The New Stack's is the story. Click through — they earned it.
Today's Five, front page: the wire scored it 6 of 10; the stamp is mine.
This lands hardest with a leader scaling past one agent. 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 3 more today: “The five components of building a "company harness"”; “AI agent security is a lifecycle — not a gatekeeper”, and 1 more.
Latent Space — 123 appearances in the Times since September 5, one of them today.
Our page is the summary; Latent Space's is the story. Click through — they earned it.
The wire scored it 5 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 3 more today: “The five components of building a "company harness"”; “AI agent security is a lifecycle — not a gatekeeper”, and 1 more.
The New Stack — 318 appearances in the Times since August 22, 4 of them today.
Our page is the summary; The New Stack's is the story. Click through — they earned it.
From our research notes:
Ollama Blog: 16 in our pages since September 9. I read them all — emphasis on investigative.
Go read Ollama Blog. They earned the byline.
No story here today, so I've got nothing in my notes.
From our research notes:
Ollama Blog: 16 in our pages since September 9. I read them all — emphasis on investigative.
Go read Ollama Blog. They earned the byline.
This desk came up empty. The front page didn't.
Not enough to call it yet. Back to the notebook.
Towards Data Science: 105 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.
Towards Data Science: 105 in our pages since September 5. I read them all — emphasis on investigative.
Go read Towards Data Science. They earned the byline.
We don't have much on this one. I'm still looking.
LangChain Blog: 61 in our pages since September 9. I read them all — emphasis on investigative.
LangChain Blog has more than I can fit in this notebook.
From our research notes:
Ollama Blog: 16 in our pages since September 9. I read them all — emphasis on investigative.
Go read Ollama Blog. They earned the byline.
Not enough to call it yet. Back to the notebook.
Towards Data Science: 105 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.
Towards Data Science: 105 in our pages since September 5. I read them all — emphasis on investigative.
Go read Towards Data Science. They earned the byline.
I have the headline. I don't have the answer yet.
The New Stack: 318 in our pages since August 22. I read them all — emphasis on investigative.
Read The New Stack's original. They did the reporting.
Something changed. I want to know what.
The New Stack: 318 in our pages since August 22. I read them all — emphasis on investigative.
This one's theirs. Read The New Stack's reporting.
We don't have much on this one. I'm still looking.
LangChain Blog: 61 in our pages since September 9. I read them all — emphasis on investigative.
LangChain Blog has more than I can fit in this notebook.
From our research notes:
The New Stack: 318 in our pages since August 22. I read them all — emphasis on investigative.
The New Stack has the full story. Start there.
Something doesn't quite line up here.
Latent Space: 123 in our pages since September 5. I read them all — emphasis on investigative.
Latent Space has more than I can fit in this notebook.
From our research notes:
The New Stack: 318 in our pages since August 22. I read them all — emphasis on investigative.
Don't stop with my notes. The New Stack has the original.