CJ Avilla

i'm happy to share some recent work on safely scaling a team of 1,000 Fable 5.2 agent to tackle ambitious, long-horizon work. the team was able to formalize a proof of Fermat's Last Theorem through coordination within our safety boundaries. read it here: claude.dev/blog/safely-scali…

check out my new article in the Anthropic eng blog. i talk about "dreaming" (offline compute) as a way to organize Claude's memory. it can correct mistakes, create memories, or build new representations. xitter.twitterapi.xyz/AnthropicAI/status/211…

Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…

We're sitting down with and @cjav_dev (@AnthropicAI) on Nov 17th at 10am PT to talk about tracing Claude Agents with OpenTelemetry and Sentry 🧠 In this live session, @Sergical and CJ will show you how to send Claude Platform OTel data to Sentry and trace a Claude agent as it calls tools alongside your existing infrastructure. RSVP to join us live 👉 sentry.io/resources/claude-o…

Your agents should sleep on it. Agents get should get smarter over night as they learn from all their previous runs! Build your own dreaming service with dynamic workflows in Claude Managed Agents. platform.claude.com/cookbook… love to hear what you think!

Message threads are now in public beta on the Messages API. The API caches your conversation on the server between turns, so instead of resending the full conversation, you send only new messages and the previous message ID, and Claude starts responding sooner.

I'm now building an "improvements" feature into all of my personal vibecoded applications: a freeform text input that's hooked up to an agent that yolo implements whatever is submitted. It's an interesting experience -- the activation energy barrier to sanding any rough edge or implementing any new feature becomes extremely low. Someone else who uses one of the applications I created has already implemented 150 improvements over the course of a few days. This pattern feels like a new kind of tool -- liquid software? -- and every application that doesn't have this feature now has a slightly rigid and anachronistic flavor. Why can't I just tell it how I want it to be different? Now, it's not obvious to me how this pattern can or should scale. Does it become unmaintainable after a while? (Maybe model intelligence improvement outstrips my feature appetite such that it's fine?) How does collaborative software work in this model? Can software designed this way be secure and reliable? Whatever about the larger questions about broader deployment, however, it definitely works great in a personal context. It reminds me a little of the visions of Lisp/Lisp machines, Smalltalk, and HyperCard -- systems that the user can take control of and modify. Those visions never quite worked, though: the vast majority of software today is not end-user modifiable in any useful way. So, while this "liquid" affordance is a very small touch, the way it changes one's relation to the software feels a bit transgressive and empowering in a Lickliderian and Engelbartian way.

We've raised the max request size for Messages and Token Counting on the Claude API from 32 MB to 128 MB. Image-heavy requests like long computer-use sessions and big PDFs were previously hitting 32 MB with plenty of context window left. Now you get ~4x more room.

17 yo never listens to us and yet hears everything we say about him, even in a whisper. I think he runs a side process using 2% of his brain listening for his name, and if he hears it, then he actually interprets the phonemes he's been buffering.

What I built with my free Claude API credits Wanted to share what I've built with the new free API credits you get with a Claude Max or Team sub. If you didn't know about these, go check out the instructions here: platform.claude.com/docs/en/… My favorite way to build your first simple no-code agent is to use our Agent quickstart on Console: platform.claude.com/workspac… Managing my schedule and everything going on in my kids life is painful, so I wanted to build an agent that reads my email for anything from my kids' schools, and then for the relevant items throws a calendar invite for me and my husband. I literally say this to Claude via the quickstart. Claude then helps me build the agent: Step 1: Claude writes the agent def with the MCPs I would need for Gmail and GCal. Step 2: Claude creates your env with the appropriate network config. You want limited here given my agent isn't searching the web. Step 3: Create your vault with credentials for Gmail and GCal. Again this is a few clicks, and we keep these refreshed for you. Step 4: Set up a deployment that runs this once a week on Sundays. Some fun things you can do: Set a budget for your sessions so you don't blow past your credits Add your partner to every invite so they can stay in the know And that's it. Now I know every event happening without having to manually add it to my calendar or remember it. Voila! Let me know if you liked this and I can share more posts like these.

Claude Projects has completely changed how I work 96% of my commits come from Projects, 4% from Routines, and 0% from the CLI I don’t think there’s any going back The CLI is great for local testing, but for prod… Claude Projects is all I need