🤖 Devs don't write code anymore. They make decisions.
Simon Willison's new guide on agentic engineering patterns is out, and it's blunt: code generation is cheap now, so your job is architecture and judgment, not keystrokes.
Greenfield beats editing. Let the agent build fresh, keep the old version as fallback. Less legacy debt, more throwaway prototypes.
Honest take from the guy who coined "prompt injection."
Simon Willison's new guide on agentic engineering patterns is out, and it's blunt: code generation is cheap now, so your job is architecture and judgment, not keystrokes.
Greenfield beats editing. Let the agent build fresh, keep the old version as fallback. Less legacy debt, more throwaway prototypes.
Honest take from the guy who coined "prompt injection."
Simon Willison’s Weblog
Writing about Agentic Engineering Patterns
I’ve started a new project to collect and document Agentic Engineering Patterns—coding practices and patterns to help get the best results out of this new era of coding agent development …
⚡️ Anonymous model drops on OpenRouter and nobody knows who built it
"Ox Alpha" appeared out of nowhere on Aug 20, free to use, no author attached. 1M-token context, video input, coding benchmark scores beating GPT-5.6. Community fingerprinting leans toward a Chinese lab (Z.ai / Zhipu GLM family), but nothing's confirmed.
Free for about a week. Someone's burning serious compute to get attention. Source
"Ox Alpha" appeared out of nowhere on Aug 20, free to use, no author attached. 1M-token context, video input, coding benchmark scores beating GPT-5.6. Community fingerprinting leans toward a Chinese lab (Z.ai / Zhipu GLM family), but nothing's confirmed.
Free for about a week. Someone's burning serious compute to get attention. Source
Business Insider
A mysterious free AI model is impressing developers. And nobody knows who made it.
Ox Alpha, a mysterious AI model, debuted on OpenRouter on Thursday. Its origins are unknown, though some suspect a Chinese lab.
⚡️ ChatGPT's tokenizer has a Japanese internet alter ego
A fried chicken question led one dev down a rabbit hole: GPT's tokenizer has surprisingly deep roots in Japanese web culture, and those training data fingerprints show up in weird, specific ways.
Non-English text was never an afterthought. It's baked into the token merges. And sometimes the model acts accordingly.
A fried chicken question led one dev down a rabbit hole: GPT's tokenizer has surprisingly deep roots in Japanese web culture, and those training data fingerprints show up in weird, specific ways.
Non-English text was never an afterthought. It's baked into the token merges. And sometimes the model acts accordingly.
Theo Steiner
ChatGPT Is a Japanese Internet Degen, or: How Fried Chicken Sent Me Down a Tokenizer Rabbit Hole
A stray piece of 2ch boilerplate in a ChatGPT answer sent me down a rabbit hole through Japanese internet history, tokenizers, training data, and one very weird token.
⚡️ AI unlocked the hard stuff
Armin Ronacher's latest post nails something real: agentic coding is making devs more ambitious.
Stuff that was gatekept (eBPF, custom crypto, DWARF files, bare-metal drivers) is suddenly approachable. Language choice barely matters now. People ship Zig and Rust code without knowing either.
More slop, sure. But maybe also more developers who actually care about fast, small, correct code.
Armin Ronacher's latest post nails something real: agentic coding is making devs more ambitious.
Stuff that was gatekept (eBPF, custom crypto, DWARF files, bare-metal drivers) is suddenly approachable. Language choice barely matters now. People ship Zig and Rust code without knowing either.
More slop, sure. But maybe also more developers who actually care about fast, small, correct code.
Armin Ronacher's Thoughts and Writings
Fast and Hard Code
People pick languages differently with LLMs.
❤1
⚡️ AI coding removes the friction that builds real skill
The argument isn't that AI writes bad code. It's that struggling through bad code is how junior devs actually learn.
Lars Faye makes the case that agents handle the exact tedious reps (debugging, syntax, boilerplate) that wire expertise into your brain. Skip those, and you're building on sand.
Calculators didn't kill math. But they did kill mental arithmetic. Same deal.
The argument isn't that AI writes bad code. It's that struggling through bad code is how junior devs actually learn.
Lars Faye makes the case that agents handle the exact tedious reps (debugging, syntax, boilerplate) that wire expertise into your brain. Skip those, and you're building on sand.
Calculators didn't kill math. But they did kill mental arithmetic. Same deal.
Larsfaye
AI Coding will Prevent Expertise | Lars Faye
The need for ongoing friction in long-term skill formation.
🚨🔥 MS Paint secretly watermarks your local AI images with a server-linked GUID
Even when generation runs on your own NPU, the prompt goes to Microsoft for moderation and the output gets stamped with a server-issued identifier, baked into both pixel data and C2PA metadata.
It's not an AI label. It's a fingerprint. The reverse engineering shows Microsoft calls this "InvisMark" and there's no off switch.
"Local" apparently just means they outsourced the compute to you.
Even when generation runs on your own NPU, the prompt goes to Microsoft for moderation and the output gets stamped with a server-issued identifier, baked into both pixel data and C2PA metadata.
It's not an AI label. It's a fingerprint. The reverse engineering shows Microsoft calls this "InvisMark" and there's no off switch.
"Local" apparently just means they outsourced the compute to you.
Xusheng Li
Microsoft Paint and Photos Embed Server-Issued GUIDs as Invisible Watermarks in Locally-Generated Images
Reverse engineering reveals how Paint and Photos embed a server-issued GUID into the pixels of locally generated AI images.
⚡️ OpenAI cuts GPT-5.6 Sol prices 20-33%
Input drops from $5 to $4 per 1M tokens. Output from $30 to $20. Valid through at least Nov 21.
Sol's still 20x pricier than Luna, but the gap vs. Anthropic just got a lot narrower.
Source
Input drops from $5 to $4 per 1M tokens. Output from $30 to $20. Valid through at least Nov 21.
Sol's still 20x pricier than Luna, but the gap vs. Anthropic just got a lot narrower.
Source
OpenAI Developers
Pricing | OpenAI API
Pricing information for the OpenAI platform.
⚡️ a16z is funding your worst timeline on purpose
A new analysis maps a16z's AI portfolio and it's not subtle. AI companions linked to teen suicides. Gambling apps dressed as "prediction markets." Tools explicitly designed to target lonely, isolated users because they pay more.
They wrote down that "high willingness to pay" thing. In a pitch doc.
A new analysis maps a16z's AI portfolio and it's not subtle. AI companions linked to teen suicides. Gambling apps dressed as "prediction markets." Tools explicitly designed to target lonely, isolated users because they pay more.
They wrote down that "high willingness to pay" thing. In a pitch doc.
www.modelrepublic.org
Andreessen Horowitz is shaping AI policy — while investing in a bleak vision of the future — Model Republic
The firm’s investment portfolio is full of companies that have exploited legal loopholes, created disturbing products, and broken the law.
🤖 Thomson Reuters spent $40M to train its own LLM and ditch the API middlemen
Meet "Thomson," TR's first in-house frontier model, built on an open-source base and trained on decades of proprietary legal, news, and tax data no one else has.
They claim fine-tuning on the right data beats plugging GPT-4 into your search bar. No public evals to verify that. Yet.
$40M for full model ownership. Ballsy call.
Meet "Thomson," TR's first in-house frontier model, built on an open-source base and trained on decades of proprietary legal, news, and tax data no one else has.
They claim fine-tuning on the right data beats plugging GPT-4 into your search bar. No public evals to verify that. Yet.
$40M for full model ownership. Ballsy call.
Thomson Reuters
Thomson Reuters Leverages its World-Class Data Assets to Launch Its Own Frontier Model
Thomson, the company's proprietary LLM, was trained and is run at a fraction of the cost of comparable frontier models and remains fully owned and controlled by Thomson Reuters
❤1
🚨🔥 vLLM had a bug that let the model own its own host machine
CVE-2025-9141: vLLM's Qwen3 Coder parser passed tool args to
Gemini flagged it as critical. Maintainer merged it anyway.
Boyd Kane's essay lays out the full attack surface. Inference engines are just software. They have bugs.
CVE-2025-9141: vLLM's Qwen3 Coder parser passed tool args to
eval(). The LLM controls the tokens sent to the inference engine, so a malicious model could emit a sequence that exploits the software loading it onto GPUs.Gemini flagged it as critical. Maintainer merged it anyway.
Boyd Kane's essay lays out the full attack surface. Inference engines are just software. They have bugs.
Boyd's Blog
LLMs could control their host machines by exploiting inference engines
| Read on LessWrong | Comment on HN | Large language models often take actions running on one computer (via an agentic harness such as Claude Code or Codex), however the LLMs’ responses to prompts are computed on a different computer with GPU access.
🧠 Apple's M5 Ultra can run 100B+ parameter models. Locally. On your desk.
512GB unified memory, 1.2TB/s bandwidth. First quad-die design Apple's ever shipped. That's not a laptop chip, that's a small inference server for $2,499.
M6 is 2nm and starts at $899. Both announced today.
512GB unified memory, 1.2TB/s bandwidth. First quad-die design Apple's ever shipped. That's not a laptop chip, that's a small inference server for $2,499.
M6 is 2nm and starts at $899. Both announced today.
Apple Newsroom
Apple introduces M6 and M5 Ultra for a big leap in performance and AI compute
Apple debuted M6 in the new Mac mini and M5 Ultra in the new Mac Studio, providing an extraordinary leap in performance and AI capabilities.
🔥1
🤖 Mac Studio M5 Ultra: 512GB unified memory, 1.2TB/s bandwidth
Apple's most powerful Mac ever just landed. M5 Ultra tops out at 512GB unified memory and 1.2TB/s bandwidth, up 50% over M3 Ultra. Cluster multiple units via Thunderbolt 5 and RDMA for a shared memory pool.
Good for running genuinely massive local LLMs. Starts at $5,499.
Apple's most powerful Mac ever just landed. M5 Ultra tops out at 512GB unified memory and 1.2TB/s bandwidth, up 50% over M3 Ultra. Cluster multiple units via Thunderbolt 5 and RDMA for a shared memory pool.
Good for running genuinely massive local LLMs. Starts at $5,499.
Apple Newsroom
Apple introduces new Mac Studio with M5 Max and M5 Ultra
Apple announced the new Mac Studio, featuring M5 Max and the all-new M5 Ultra, delivering a monumental leap in AI performance and even faster graphics.
🔥3
⚡️ Alibaba's next model: 125B params, only 6B active per token
Qwen3.8-Flash-Next just dropped on ModelScope. MoE beast: 125B main params plus 51B N-gram embeddings, but activates just 6B per token. Matches Qwen3.7-Plus at roughly 1/9th training cost.
It's also a preview of the full Qwen4 architecture. Sparse attention, new MoE design, released early so the community can prep.
Qwen3.8-Flash-Next just dropped on ModelScope. MoE beast: 125B main params plus 51B N-gram embeddings, but activates just 6B per token. Matches Qwen3.7-Plus at roughly 1/9th training cost.
It's also a preview of the full Qwen4 architecture. Sparse attention, new MoE design, released early so the community can prep.
www.modelscope.cn
Qwen3.8-Flash-Next - ModelScope 魔搭社区
Qwen3.8-Flash-Next - Qwen 在 ModelScope 开源的模型。
❤2
⚡️ Entry-level workers are Gen Z's canary in the coal mine
Stanford just confirmed it: AI-exposed roles for 22-25 year-olds are down 13% since 2022. Software dev is off nearly 20% since ChatGPT launched.
No juniors trained today means no seniors tomorrow. At some point that scarcity flips costs and forces companies to hire juniors again. Cold comfort for whoever graduates in the meantime.
Stanford just confirmed it: AI-exposed roles for 22-25 year-olds are down 13% since 2022. Software dev is off nearly 20% since ChatGPT launched.
No juniors trained today means no seniors tomorrow. At some point that scarcity flips costs and forces companies to hire juniors again. Cold comfort for whoever graduates in the meantime.
Ars Technica
AI is hitting entry-level jobs hardest, Stanford study finds
Young employment in AI-impacted fields down 19% compared to more AI-resistant occupations.
❤2
⚡️ OpenAI's Jalapeño chip beats Nvidia Blackwell on perf/W
SemiAnalysis ran OpenAI's in-house inference silicon (built with Broadcom) against Blackwell in their InferenceX benchmark. Jalapeño wins across almost all scenarios, low-latency AND high-throughput, without being tuned for any specific workload.
So OpenAI's now a chip company. Nvidia's watching.
SemiAnalysis ran OpenAI's in-house inference silicon (built with Broadcom) against Blackwell in their InferenceX benchmark. Jalapeño wins across almost all scenarios, low-latency AND high-throughput, without being tuned for any specific workload.
So OpenAI's now a chip company. Nvidia's watching.
Semianalysis
OpenAI Jalapeño: Better Than Nvidia Blackwell
OpenAI’s self-designed ASIC compared with Rubin, Jalapeño’s TCO, throughput per MW, and spicy deets
❤2
⚡️ OpenAI's data center chief lasted less than 6 months
Chris Malone, who oversaw OpenAI's data-center buildout, left last week, per WSJ. He joined in March 2025, right after Stargate dropped, coming from Meta where he led data-center strategy.
Barely into vesting. Stargate still being built. Make it make sense.
Chris Malone, who oversaw OpenAI's data-center buildout, left last week, per WSJ. He joined in March 2025, right after Stargate dropped, coming from Meta where he led data-center strategy.
Barely into vesting. Stargate still being built. Make it make sense.
The Wall Street Journal
Exclusive | OpenAI’s Head of Data Centers Has Left the Company
Malone joins a string of recent high-level executive departures as the AI giant heads toward an IPO and ramps up its spending on computing power.
❤4
🤖 Your agent isn't dumb. It's drowning.
New paper argues most production agent failures aren't reasoning failures. They're context failures: histories and tool outputs bloating the window every turn until the agent loses track of what it was doing.
Memory isn't a storage problem. It's a lifecycle one. What to remember, when to compress, when to forget are architecture decisions, not afterthoughts.
Source
New paper argues most production agent failures aren't reasoning failures. They're context failures: histories and tool outputs bloating the window every turn until the agent loses track of what it was doing.
Memory isn't a storage problem. It's a lifecycle one. What to remember, when to compress, when to forget are architecture decisions, not afterthoughts.
Source
arXiv.org
Agentic Context Management: Solving Agent Memory and Cost by...
Production AI agents' failures are less often due to an inability to reason well and more often because they cannot manage what is in their reasoning context: conversation histories, large...
🚨🔥 First confirmed AI autonomous drone kill. Nvidia chip inside.
A Russian Molniya drone chose its own target and hit a gas station in Zaporizhzhia on July 6, killing three civilians with no human pilot issuing the final command.
An Nvidia Jetson Orin did the targeting. It was found in the wreckage, still legible under the soot.
Researchers call it the first documented case of civilian deaths from a fully autonomous Russian drone. That line just got crossed. Source
A Russian Molniya drone chose its own target and hit a gas station in Zaporizhzhia on July 6, killing three civilians with no human pilot issuing the final command.
An Nvidia Jetson Orin did the targeting. It was found in the wreckage, still legible under the soot.
Researchers call it the first documented case of civilian deaths from a fully autonomous Russian drone. That line just got crossed. Source
Tom's Hardware
Nvidia Jetson Orin-guided Russian AI drone killed three civilians in Ukraine, forensic teams say — first documented case of civilian…
The unencrypted module let Ukrainian investigators read the drone's terrain maps and targeting code.
❤2
⚡️ Manual coding is going extinct, says InfluxDB founder
Paul Dix argues we've hit the inflection point. Bun 1.4's million-line Rust rewrite? Done almost entirely by AI agents. And those models weren't even frontier-level.
He thinks humans will soon review only the output, not the code. Programming doesn't die, it just goes the way of typesetting.
Reasonable take or cope? The GitHub commit graphs are hard to argue with.
Paul Dix argues we've hit the inflection point. Bun 1.4's million-line Rust rewrite? Done almost entirely by AI agents. And those models weren't even frontier-level.
He thinks humans will soon review only the output, not the code. Programming doesn't die, it just goes the way of typesetting.
Reasonable take or cope? The GitHub commit graphs are hard to argue with.
Pauldix
The end of programming — Paul Dix
Creating software by writing code by hand and having humans review it is coming to an end. Why the Bun 1.4 release marks the end of programming as we've known it.
❤2
🤖 Russian ops used ChatGPT to fake Western academics. OpenAI just caught them.
The campaign ran a site pushing plagiarized research and a made-up "sovereignty index" that conveniently ranked Putin's Russia on top. Comments seeded on Facebook, Telegram, and Substack.
How they got caught: prompts were in Russian, but outputs were in English. The model leaked gendered grammar ("Germany... she") straight from Russian syntax. A very human giveaway inside a machine-made text.
The campaign ran a site pushing plagiarized research and a made-up "sovereignty index" that conveniently ranked Putin's Russia on top. Comments seeded on Facebook, Telegram, and Substack.
How they got caught: prompts were in Russian, but outputs were in English. The model leaked gendered grammar ("Germany... she") straight from Russian syntax. A very human giveaway inside a machine-made text.
OpenAI
Disrupting a new covert influence campaign from Russia
OpenAI banned Russia-origin accounts using AI to promote a fake Israel-based think tank and a “sovereignty” index praising Russia and criticizing the West.
❤1