If you're like me, you had no idea there was a waitlist to sign up for to get access to Claude Projects. So you and I both missed out on getting access this morning 🤦‍♂️ Add yourself to the waitlist here:

Huge differentiator for Grok @bot here because it'll give you access to search, read and monitor @X for free. So one thing you can do is use @mvanhorn's Agent Tincan project ( to get your other agents use your grok @bot for X functions if you prefer to use those day to day while not missing out on this.

A few months ago we introduced a prototyping skill into Compound Engineering, and last month @kieranklaassen and I teamed up on it to improve it based on our own usage. The biggest change is live annotation and an iteration loop with your agent. Here's a quick demo showing a basic loop.

Look at these results on @VulcanBench by @morganlinton on GPT-6.1 Sol are various effort levels. It's the only consistent benchmark doing the unglamorous work to give us this level of detailed insight that I'm aware of. 6.1 Sol at Low looks incredible, and if you combine it with multi-model code reviews for example (which compound engineering does), you can really get great results while pinching pennies here. I'm gonna try it on fast mode too.

Most Jev demos get shared in a vacuum, focused on cost and speed. But mysteriously absent is any discussion about comparative quality against an LLM driving for the same outcome. Yesterday I ran Jev in a re-ranking pipeline with an eval harness. It was ~50× cheaper and ~25× faster than the LLM path. We still kept Opus. Why? Opus produced better results on that task. We’ll pay more money and more time when the ranking is better. That shouldn’t be controversial — you just never hear it next to the flashy Jev demos. The interesting part of Jev isn’t “replace your LLM.” It’s the new loops you can afford when a decision is insanely cheap, and insanely fast.

You’re probably eligible for the Apple Siri settlement. Ask your @muse @bot or Instinct: Find the info about the Apple Siri settlement. Then help me file my claim. Check my eligibility, list what info I need to give you that you don’t have, and fill out the claim form for me. Double check any info with me you are unsure about.

Just updated @illo_skill to work with @muse using its built-in image gen tools. To install, just tell your muse: --- install the illo skill from tmchow/illo-skill github repo and use the skill to install all the characters for me. Periodically check and keep my skill and character packs up to date. --- You can then create a persistent side chat called "illo" or "Assets" and have all your illo images in one chat. 🔗 🤓

It's exhausting reading some reactions to @bot, @muse, and Instinct’s success with “but ChatGPT and Claude can already do that”. The innovation isn’t intelligence, or even capability for most of this stuff. It’s the packaging. Personal AI bots make two things ridiculously simple: 1. knowing where to start by default. no dance of opening a new chat, then hunting the old thread inside a project 2. finding an older discussion without it turning into a nightmare No surprise Grok Build exists alongside Grok Bot. @SpaceXAI didn’t say “we have Grok Build, why do we need Grok Bot?” Sure, ChatGPT and Claude both give you a giant text box when you start. But if you want the discussion you started three hours ago? You’re sifting through a junk drawer of threads.

1. Earlier this year, Amazon was caught scraping and doing delegated buying with their “buy for me” program: 2. Why is Amazon cutting off a source of revenue? muse and other agents buying on behalf of users is a good thing for them. 3. If you setup @Muse with @Tailscale (there’s a connector) you may be able to bypass this unless Muse uses a specific user agent with their browser.

Whenever new site publishing hosts get popular, first thing I do is to check what is indexed in Google. Yep, Google has been crawling the sites. Most aren’t accessible but ~10% are. Even just scanning the titles tells you about what people are building.

While @typesafeai Jev is now available without a waitlist, the better move is to just use it through the @vercel AI Gateway since it's free through 9/25. Sure Jev is cheap but going through vercel gives you some very easy ways to chain models together along with jev.

I love the ergonomics of the @KeychronMK Q11. I bought a second one but hated the KSA keycaps so much they were too tall. Snagged these new CSA profile ones which are so much shorter and make typing feel so much better.