Related posts

Author and clean-energy analyst who writes on solar cost curves.

• This matches my expectations. Cheaper models (open and closed) will capture most of the token volume. But in a great many domains (and anything adversarial, zero sum, or first past the post), people will pay for the very most advanced model.
  Quoting@KonstantinPilzNarrative violation: Open models still account for only 10% of global spending on AI inference.

  Related

CEO of Vercel; creator of Next.js and Socket.IO. Writes at rauchg.com.

• Looks like today may be a record day for token volume % of open models on Vercel AI Gateway: 🟦 Open 78.4% 🟨 Closed 21.6% While spend 💲 usually tells a different story, #3 and #4 today are Moonshot AI & DeepSeek. Adding Z⁠.ai, their combined spend surpasses OpenAI (#2). (Do note that's the spend for inference of the model across providers (mostly in the US), not revenue going directly to the open weight labs.)

  OpenAIRelated

Former L8 engineer at Meta, Microsoft and Atlassian, now writing and building solo on agentic engineering. Writes Kun's Field Notes and posts a lot about AI coding agents on X.

• axi — Design principles for agent ergonomics. Higher accuracy with lower token cost than both MCP and regular CLI.

  MCPRelated

Creator of Ruby on Rails, CTO of 37signals, and creator of Omarchy.

• Great to welcome @OrcaRouter as a Distinguished Patron with a $150,000 donation to the Strategic Token Reserves for Omarchy! We're going to put their new OrcaCyber Zero 1.0 model to great use with our security and kernel work 🤘

  omarchy.org

  OmarchyRelated

Writes The Pragmatic Engineer, the software-engineering newsletter, and wrote The Software Engineer's Guidebook. Formerly an engineering manager at Uber, and at Microsoft/Skype and Skyscanner before that.

• Here's what OpenAI's agentic software factory looks like, today. It's surely not token efficient, but eg agents "babysitting" releases to prod is something new, and an interesting approach. Details:

  newsletter.pragmaticengineer.com

  OpenAIRelated

Co-founder of Modem and a founding engineer at Sentry, where he went on to be VP of Engineering. Co-author of Third-party JavaScript, and co-host of the State of Agentic Coding podcast with Armin Ronacher.

• The industry accepts there are 10x engineers But it hasn’t yet accepted there are 10x engineering *prompters* IMO it’s a real thing - and I’m not talking software factory token maxxers

  Related

Moroccan senior frontend developer writing at smakosh.com; builds side projects and client work under Smakosh LLC.

• Token breakdown improved on the analytics page on @llmgateway

  Related

Austrian developer. Created the libGDX game framework, and more recently the pi coding agent, which moved with them to Earendil — Armin Ronacher's company — in April 2026. Writes at mariozechner.at.

• look at me. i'm a token seller now. (please give us money, i need it for more icecream)
  Quoting@mitsuhikoIf you want to play around with an early thing: you can now give us money and we give you tokens in return.

  Related

Former L8 engineer at Meta, Microsoft and Atlassian, now writing and building solo on agentic engineering. Writes Kun's Field Notes and posts a lot about AI coding agents on X.

• ok everyone, i took one for the team here's the data we all wanted to see - real token value of each LLM subscription, empirically measured through usage on my real subscriptions - supergrok heavy has now become the highest value at $12k worth of tokens (40x ROI) - the $200 plan from openai and anthropic roughly give the same amount of token value, $7k give or take - cursor ultra's ROI is the lowest among mainstream subscriptions, roughly half of OAI or Ant. for individual consumers i'm afraid this no longer makes sense to purchase this was done in a clean environment. i ran each provider acr…

  LLMsAnthropicRelated

Latvian full-stack WordPress developer and course creator at WPElevator; blogs since 2007 about the open web, electronics, home automation and sustainable living.

• Today I learned that @OpenAIDevs MCP client https://t.co/nhRPF6KVEf lists private_key_jwt for the token endpoint auth while the CIMD endpoint is actually public (auth=none). So your MCP OAuth must be able to skip private_key_jwt, if unsupported.

  MCPRelated

Co-creator of Django and creator of Datasette; writes daily at simonwillison.net.

• Interesting details in here about how OpenAI researchers are using coding agents I'd love to know what caused the huge uptick in token spend in mid-July, was that when Astra became available to employees?
  Quoting@kliu128Today we're releasing data on models accelerating research at OpenAI. Recursive self-improvement could be the most important contributor to AI capabilities over the next few years, but by default it will only be seen inside a few frontier AI labs. Being transparent is more urgent than ever, so we c…

  coding agentsOpenAIRelated

Venture capitalist and founder of Theory Ventures, previously a partner at Redpoint. Writes a near-daily blog on startups, SaaS metrics, data and AI.

• How large is Snowflake's AI revenue? Management lowered product gross margin guidance from 76% → 74% due to "fast-growing AI workloads." If GPU inference runs at ~25%–35% gross margins, that 200 bps compression implies ~$60M–$75M/quarter (~$240M–$300M ARR) in direct AI token consumption.

  Related

Engineer at Vercel Labs as of 2026. Ships a steady stream of small developer tools — agent-browser, portless, json-render, native-sdk, the v0 SDK — and lists every one of them at ctate.dev.

• One more for the night: emulate v0.11 – Shared Next.js/Nuxt adapter runtime with persistent GitHub App key state – Linear issues now expose priorityLabel – GitHub token inspection adds perms, repo access, expiry and status

  github.com

  Related

Chilean web engineer building web experiences since 2006; co-founder of Media Creators and writes at iolivares.com.

• La suscripción de 5x viene rindiendo, pero no estoy seguro si con el cambio del -17.5% de sesión vaya a estar bien. Pensando en probar Codex, veo a todos disfrutando de esa fiesta. Estamos en un mundo donde estar Token Maxing con una suscripción dejó de ser suficiente, ahora necesito 3 (Claude, Cursor y al parecer también Codex). 💸

  English, from SpanishThe 5x subscription is paying off, but I'm not sure if with the change of the -17.5% session it will be okay. Thinking about trying Codex, I see everyone enjoying that party. We're in a world where Token Maxing with one subscription is no longer enough, now I need 3 (Claude, Cursor, and apparently Codex too). 💸

  AnthropicRelated

Co-founder and CEO of Stripe and co-founder of the Arc Institute. Keeps a personal site of reading lists, open questions and notes on scientific progress and how research is funded.

• Between wallets for agents with @link (and @privy), MPP, Tempo, MCP, Stripe CLI, Agentic Commerce Suite, sandboxes, OpenRouter, Metronome/metered billing, Radar and distillation/token fraud, it turns out that "the economic infrastructure for AI" implicates a lot. Pretty much every layer of the Stripe stack is changing, and it’s a very good time to join if you want to build high-scale software that facilitates the practical deployment of this new intelligence.

  MCPRelated

Venture capitalist and founder of Theory Ventures, previously a partner at Redpoint. Writes a near-daily blog on startups, SaaS metrics, data and AI.

• Does the token supply chain also experience the bullwhip effect? Here's my explanation of this classic phenomenon and what it means for AI data centers.

  data centersRelated

AI research engineer working on large language models. He writes the Ahead of AI newsletter and is the author of Build a Large Language Model (From Scratch).

• x.comA couple of days ago, I did a quick explainer on Claude’s new watermarking process and implementation. Since it’s such a popular topic and sparked such a lively discussion, I thought it might be interesting to go into a bit more detail when explaining how it works. So, instead of the usual text article, I recorded a little lecture on the topic (to change it up a bit from my usual articles). It ended up a bit longer than intended, but I hope it clarifies a lot of things: - Sampling the next token in an LLM and pseudorandom number generators - How watermarking relates to the regular LLM samplin…

  LLMsAnthropicRelated

• NewsletterHow Claude Watermarks AI-Generated Text

  A 48-minute video walkthrough of token sampling, watermark detection, and removal

  AnthropicRelated

Indonesian software engineer, educator and engineering manager at Zero One Group; open-source enthusiast writing at ripandis.com.

• DeepSeek V4 Flash naik harga. Limit usage di OpenCode Go juga turun drastis. Untung masih ada 17 juta (17M) token per hari gratis. Mau? Daftar sini aja:

  English, from IndonesianDeepSeek V4 Flash price increased. Usage limit on OpenCode Go also dropped drastically. Luckily there are still 17 million (17M) tokens per day for free. Want some? Register here:

  s.id

  Related

AI research engineer working on large language models. He writes the Ahead of AI newsletter and is the author of Build a Large Language Model (From Scratch).

• A short illustration of how the Claude's watermarking is supposed to work (based on my read of their released materials). In general, when we are generating tokens, there can be multiple high-scoring tokens at certain next-word positions. Usually, we sample with top-k or top-p sampling so the highest-scoring token is most often selected (if we repeat the sampling many times), but other tokens may be selected as well. With watermarking, there is a key that says which of the (ideally equally) highest-scoring tokens to select. Or, more concretely, the secret key and previous token influence the…
  Quoting@AnthropicAIWe’ve written an FAQ to answer some of the questions we've received about watermarking. In summary: • We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking; • Our watermarking metho…

  AnthropicRelated

Security researcher; built Report URI and Security Headers, and writes about web security in practice.

• Device Bound Session Credentials lands in Chrome on macOS

  Device Bound Session Credentials (DBSC) is Chrome's answer to session cookie theft, usually by InfoStealer malware. Instead of a cookie being a bearer token that works anywhere it's pasted, DBSC binds the session to a p…

  Related

Programmer and writer on computer architecture, performance, and software reliability. He has worked on CPU design at Centaur Technology and on software at Google and Microsoft, and writes long-form technical essays at danluu.com.

• How do programming languages impact token efficiency and correctness?

  danluu.com

  Related

Canadian full-stack web developer who makes JavaScript and CSS video courses and co-hosts the Syntax podcast.

• I bought $100 of Claude API credits for $8 on the Chinese black market It's an insane, well run industry — some losing $10 million per month to theft Join me as I go deep on the AI token underworld

  youtube.com

  AnthropicChinaRelated

Full-stack developer, creator of Tailwind CSS; writes at adamwathan.me and makes courses and screencasts for web developers.

• Feels like we’re about 6 months away from a token is a token is a token. Just gas at the gas station.

  Related

Founding member of OpenAI and former director of AI at Tesla; creator of nanoGPT and the term "vibe coding".

• We're starting to leave the territory where you'd test an LLM by e.g. "create an svg of pelican on a bicycle". As one idea to generalize it, I was interested what Opus 5 would do if I gave it the first paragraph of the Lord of the Rings, a 1M token budget (~$10) and asked for three js render of it. Opus went off for ~2 hours and wrote 5500 lines of code that (procedurally) rendered the story. It's kind of janky but fun. But it's a bit mindboggling that the LLM has to place and orchestrate various polygon assets in (x,y,z) coordinates and write code that animates it all, and that it even does…

  LLMsRelated

Software design writer and chief scientist at Thoughtworks; wrote Refactoring and Patterns of Enterprise Application Architecture.

• The Economic Benefit of Refactoring

  Giles Edwards-Alexander does an experiment to see if decomposing a large function helps reduce token costs, suggesting that is may now be possible to measure the economic benefit of refactoring more

  Related

Technologist and entrepreneur; former CTO of Threadless and chief technology officer of the 2012 Obama re-election campaign.

• "Sorry for the token bonfire, Doctor Biz."

  Related

Software design writer and chief scientist at Thoughtworks; wrote Refactoring and Patterns of Enterprise Application Architecture.

• The Orchestrator's Tax

  Subagents get justified by time saved and parallel execution, but Rahul Garg explains that's not what matters most. Every token in the orchestrator's context is competing for its attention, and the real value of a subag…

  Related

Co-founder of Modem and a founding engineer at Sentry, where he went on to be VP of Engineering. Co-author of Third-party JavaScript, and co-host of the State of Agentic Coding podcast with Armin Ronacher.

• 👋 New blog post: "How coding agents read files – and how to write for them" In it I explain how code optimized for agent-discovery can lead to major token savings on search + retrieval, even with frontier models like GPT-5.6 Sol Plus other important findings ⬇️

  coding agentsOpenAIRelated

Systems and developer-tools engineer at Cloudflare as of 2026, on durable infrastructure for AI agents; worked on React and PartyKit before that. Writes at sunilpai.dev.

• never waste a token

  durable inference: resumable streams, crash recovery, and why the LLM request shouldn't die with your process.

  LLMsRelated

Co-founder of Modem and a founding engineer at Sentry, where he went on to be VP of Engineering. Co-author of Third-party JavaScript, and co-host of the State of Agentic Coding podcast with Armin Ronacher.

• 🎨 hunk v0.15.0 is out: - supports git --color-moved - transparent backgrounds - color token fixes - more themes - Sapling support (it's a VCS from Meta) also now there's a website →

  hunk.dev

  Related

Technologist and entrepreneur; former CTO of Threadless and chief technology officer of the 2012 Obama re-election campaign.

• If you wanted to try out fable but are scared of blowing away your token budget do we have a skill for you! Its called Thrifty. It makes things cheaper. We built thrifty for claude code around a better split: strong model plans → cheap model executes → gate verifies.

  AnthropicRelated

Member of technical staff at Anthropic. He has led ML/AI teams at Amazon, Alibaba and Lazada, and writes about LLMs, recommender systems and engineering at eugeneyan.com.

• Mythos evals from XBOW and UK AISI: • UK AISI: Mythos completed a 32-step network attack (est. at ~20 hrs for experts) in 6/10 tries. First model to solve their e2e cyber ranges! • XBOW: "token-for-token, unprecedented precision" Read more: https://t.co/hD6G6DvALg,

  xbow.com

  Related

Programmer and writer on computer architecture, performance, and software reliability. He has worked on CPU design at Centaur Technology and on software at Google and Microsoft, and writes long-form technical essays at danluu.com.

• What do people like for token-intensive workflows for personal projects? With a semi-autonomous workflow, I can exhaust a week's quota on the $200/mo codex plan in ~10 minutes of my time. If I'm in the loop, I get ~10x(?) the productivity per token, but it's time inefficient and

  Related

Software engineer, Wikipedia editor and crypto critic; writes Citation Needed and runs Web3 is Going Just Great.

• President Trump’s $TRUMP memecoin is preparing to launch a “Coin Club” membership scheme

  The website promises "elite and extraordinary experiences" as part of the newest scheme to revive a token that’s down 97% from its peak and still falling.

  Related

Co-founder of Modem and a founding engineer at Sentry, where he went on to be VP of Engineering. Co-author of Third-party JavaScript, and co-host of the State of Agentic Coding podcast with Armin Ronacher.

• State of Agentic Coding #5 with @mitsuhiko.at is live. In this episode: - death of the IDE premature? - what's with all the bugs - will future models fix yesterday's AI code? - token abuse, slowing down, and more! YT: www.youtube.com/watch?v=qCpY... Spotify:

  open.spotify.com

  Related

Founding member of OpenAI and former director of AI at Tesla; creator of nanoGPT and the term "vibe coding".

• LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki…

  LLMsRelated

Independent game developer, creator of the colony simulation RimWorld and author of Designing Games: A Guide to Engineering Experiences. Previously worked on BioShock Infinite.

• Why do people say AI "just" predicts the next token as though this is somehow simpler than thinking? If you can predict what someone will say, you must be tracing the same processes they used to generate those words. You must recreate their thought process. You must think.

  Related