Hello,
World News AI lands twice a week, Tuesday and Friday, in four minutes: what shipped in AI and what you can actually try the same day.
Who it's for: people building agents, from the simplest projects to the ones that push models hard, and people using AI seriously at work.
What sets us apart fits in one line: we don't reprint announcements. For every item, one sentence of fact, one sentence of consequence, and the link to the original document. When we don't know, we say so.
Here is a full issue, with the six sections that will come back every time.
What it's for the small tasks you run a thousand times a day, where every second and every cent counts, and reading images alongside text.
The counters 208,000 downloads, 4,580 likes, plus 431,000 downloads of the light build.
The limit the licence is the vendor's own, not Apache: read it before you make money with it. No image generation.
👉 huggingface.co/Qwen/Qwen3.8-Flash-Next, and unsloth/Qwen3.8-Flash-Next-GGUF for the build that runs on your own machine, with nothing leaving it.
What it's for carrying a long project without losing the thread, with documents and screenshots mixed together.
The counters 441,000 downloads, 1,849 likes.
The limit no clear licence on the model card, and the documentation is mostly in Chinese. Too heavy for an ordinary laptop.
What it's for turning a sentence or a picture into video, without the wait.
The counters 5.5 million downloads of the open weights.
The limit the model is free but heavy, and the service that streams it live is paid per use.
JordyZomer/lemmalog — 242 stars, MIT licence, last commit 27 August 2026 Memory for agents that keeps track of where each fact came from. Ships an MCP server, so it plugs straight into your agent.
What it does not do one node only, no distribution, and memory is filled by hand: nothing is extracted from conversations on its own.
Few stars, young project: that is normal and says nothing about quality. The risk here is not a trap, it is abandonment in three months.
👉 github.com/JordyZomer/lemmalog
XiaoDuoYa/codex-with-chatgpt — 2,072 stars, MIT licence, last commit 28 August 2026 ChatGPT thinks and splits the task, Codex executes.
What it does not do nothing offline, and it leans on two paid subscriptions. If either provider closes the tap, the whole rig stops. That just happened to Cursor, see below.
👉 github.com/XiaoDuoYa/codex-with-chatgpt
earthtojake/text-to-cad — 14,101 stars, MIT licence, last commit 1 September 2026 Describe a part in plain English, get the computer-aided design file back.
What it does not do no viewer, you need external CAD software to open the result, and it does not handle multi-part assemblies.
👉 github.com/earthtojake/text-to-cad
A warning that holds for the whole section. Stars can be bought, a few cents each. A project with fifty thousand stars, twelve forks and no discussion does not happen naturally. None of these tests proves anything, they only say where to look. The one protection that does not depend on your instinct is limiting what the module is allowed to do.
OpenAI cut Cursor off from its models. Developers routing through it lost their engine without warning. For every piece of your chain, write down the fallback and how long the switch takes.
ChatGPT falls under the EU's strictest regime. The Commission classes it among the very large platforms. This is the template for what will apply to the others as soon as they cross the audience thresholds.
Sony Music and Warner Chappell sue Anthropic. Filed in California, $150,000 claimed per work. After books and the press, music opens a third front, and legal risk ends up in the price of models.
An agent that fails one step in twenty fails two thirds of its tasks. That is not bad luck: a long task has to get every step right, and small errors multiply instead of adding up.
Let's keep one concrete case to the end: an agent that processes invoices arriving by email. Read the message, open the attachment, pull the amount, find the supplier, convert the currency, post the entry, acknowledge receipt. Around twenty steps. Out of thirty invoices this month, eleven make it through.
Switching models will not help, and you have no code to write. There are two kinds of failure, and confusing them costs days. Random failure trips sometimes, for different reasons: an automatic retry repairs it, and your eleven invoices become twenty-eight.
Systematic failure trips always, at the same step and in the same way: retrying changes nothing, you have to split the step in two or take it out of the model.
The move, today: open your agent's log and look at whether the same failure always comes back at the same place. That is the whole difference between an hour's repair and a lost week.
👉 The four prompts that find where your agent fails, to paste one after the other into your assistant. They list the real steps, add a log, run the task ten times in dry-run mode, and hand back a table that separates the two kinds of failure.
Which words, present or missing from an instruction, actually change what the model does. Measured, not guessed. Read it before you write your next system prompt.
Simon Willison untangles OpenAI's most confusing product, which is really two products. Useful if your company is adopting it.
An agent should not grade its own homework: models from the same vendor share blind spots. The paper proposes having the work checked by a model from somewhere else.
The trap. A coding assistant sees everything in the folder you launched it from,
and in every subfolder. Launched from your home folder or your Desktop, it has your tax returns, your contracts and your password exports. The path is displayed at the top of its window, and nobody ever looks at it.
The right move, five minutes. Open that folder and ask yourself one question: would I send its entire contents to a stranger? If the answer is no, close it, create a folder holding only the project at hand, and relaunch from there. One folder per project, never your account root, never the Desktop.
### ⚛️ Quantum insert — your data is already stolen
Why this concerns you: your agent transmits and logs data that will have to stay secret long after your project ends.
An attacker needs no quantum computer today. Recording encrypted traffic now and shelving it until the day it can be opened is enough. Today's encryption protects against today's reading, not against 2032's.
One question only: does the data my agent handles have to stay secret for more than ten years? Medical records, contracts, sources, industrial property. If yes, the subject is present tense, and resistant algorithms are already standardised. If no, sleep well on this particular point.
Quantum will never have a fixed section here. It shows up only as it does now: when it changes something about what you are building.
Every Tuesday and every Friday, four minutes, these six sections: releases, repos, in brief, the problem of the week, worth reading, the security reflex. Sometimes an insert, when an outside subject genuinely changes something for you.
Four promises that will not move: nothing that cannot be traced to its original document, an insert justified in one sentence or no insert, a sponsor always
labelled and never covered editorially, and no content hidden behind a click. What the subject line promises is in the email.
The most useful thing you can do today: forward this issue to one person who is building something. That is how this newsletter grows, or doesn't.
Reply to this email and tell us what you are building. Every reply is read.
— World News AI
The unsubscribe link is right below, in plain sight. We will never hide it in light grey on a white background.
Twice a week, Tuesday and Friday, in four minutes: what shipped in AI and what you can actually try the same day.
Double opt-in by email. One-click unsubscribe, link in plain sight in every issue.
Geneva, Switzerland. Write to hello@newsworldai.xyz.