CJ Avilla @cjavilla.bsky.social · 15huniversal malleability is coming... I think x.com/i/article/2106087767864455168 000
CJ Avilla @cjavilla.bsky.social · 16hBuilt a few useful little Claude Code mods for myself and wanted to share... 🧵 github.com/cjavdev/mods 000
CJ Avilla @cjavilla.bsky.social · 01/10/2026If you ran Sonnet 5 with `thinking: {"type": "disabled"}` and switched to Sonnet 5.5, that request now returns 400. Updated the error message today: To turn thinking off on this model, send "thinking": {"type": "between_tools"} instead of... x.com/cjav_dev/status/2105047310107… 000
CJ Avilla @cjavilla.bsky.social · 01/10/2026Couple things to note about Sonnet 5.5! `thinking: {"type": "disabled"}` returns a 400 on Claude Sonnet 5.5. The replacement is `{"type": "between_tools"}` Turns off up-front thinking. Between tool calls the model can still write short progress... x.com/cjav_dev/status/2104682336713… 000
CJ Avilla @cjavilla.bsky.social · 01/10/2026We've added inline tool definitions to the Claude API! Editing the top-level `tools` array mid-conversation changes the prefix so thinking blocks are ignored with preserved thinking which is enabled for new models. Add full tools mid-conv instead: 000
CJ Avilla @cjavilla.bsky.social · 01/10/2026New channel for us to share tips, tricks, and updates for developers: claude.dev 000
CJ Avilla @cjavilla.bsky.social · 01/10/2026Really like the idea of one-bit comms - Reactji to a Slack message kicks off Claude Tag - Hot keys for an inbox to dismiss, research, reply to have a Claude Managed Agent handle - Claude Code asking you for 1, 2, 3 in AskUserQuestion (more than a bit, but ~1 ascii-char) 000
CJ Avilla @cjavilla.bsky.social · 18/09/2026Great use-case for claude managed agents! www.cockroachlabs.com/blog/continuu… 000
CJ Avilla @cjavilla.bsky.social · 16/09/2026New machine? brew install anthropics/tap/ant Using Claude Code to drive the ant CLI to setup and use the API is so nice. 020
CJ Avilla @cjavilla.bsky.social · 16/09/2026Tokens aren't fungible. Same 600K token budget, same financial analysis bench: plain execution hit 76% accuracy. Execution plus an advisor hit 89%. Watch @katelyn_lesse and @angjiang walk through advising, grading, and dreaming as jobs for tokens. www.youtube.com/watch?v=PXj0p_mW9nI 000
CJ Avilla @cjavilla.bsky.social · 08/09/2026Playing this new mental game like code golf or vim golf where I try to see how much I can get Claude to build in a single session without compaction. The only constraint is the context window size. 130
CJ Avilla @cjavilla.bsky.social · 04/09/2026An infrastructure issue is causing a partial outage across Claude Code, the API, and claude.ai. We're working on it but don't have an ETA yet. We'll post here when that changes, and quickest place for updates is always status.claude.com 000
CJ Avilla @cjavilla.bsky.social · 02/09/2026Fable 5.1 thinking block visibility when switching models mid conversation. In this example Fable 5.1 -> Opus -> Fable 5.1 (leave the thinking blocks and they are ignored during Opus turns and then picked back up when you go back to Fable) 000
CJ Avilla @cjavilla.bsky.social · 02/09/2026Fable 5.1 mid-conversation effort changes 1. you can change the top level effort (and keep thinking blocks), but that resets the prompt cache 2. prefer changing effort mid-conversation with system messages using output_config.effort because that will not reset the cache($) 000
CJ Avilla @cjavilla.bsky.social · 01/09/2026🫰tip Run `/compact` before you step away. That way the compaction runs when the prompt cache is hot... On a ∼147k-token session with Claude Opus 5, the /compact request's input side cost $0.08 while the cache was warm vs $0.92 after it expired. 130
CJ Avilla @cjavilla.bsky.social · 26/08/2026Personal agent setup and personal website now both have embedded Claude Managed Agents with access to the codebase so I can ask them to edit themselves and push PRs that I can review. Still a missing piece I'm excited to share more about it in a blog post I'm cooking up. 🐌 010
CJ Avilla @cjavilla.bsky.social · 26/08/2026Anthropic Python SDK v1.0 is out! ‼️HTTP layer moved from `httpx` to `httpx2` github.com/anthropics/anthropic-sdk… Use `/claude-api upgrade` in Claude Code 010
CJ Avilla @cjavilla.bsky.social · 26/08/2026Computer use is GA on the Claude API. Stellar production writeup by @davilocatelli13 at Asteroid: agents driving EHRs and payer portals, human steering down from 8.9% of runs to 0% since April. asteroid.ai/blog/computer-use-in-he… 000
CJ Avilla @cjavilla.bsky.social · 26/08/2026Workbench is out, Playground is in! 🛝 My favorite feature is being able to toggle to see the code for an API call (advisor mode, tool calls, structured outputs, etc.) with all of the supported SDKs, then edit the code and see that update the form ✨ fun way to iterate Play around today 🤽 010
CJ Avilla @cjavilla.bsky.social · 19/08/2026If you previously had CVP and your application is now "In Review" a few things to try: Submit any documentation thats required. If you're still seeing in-review and unable to submit docs for some reason, please DM me your org ID so I can share with the team. 101
Reposted by CJ AvillaSan Francisco Ruby Conference, Nov 10-12, 2026 @sfruby.com · 19/08/2026Welcome @cjavilla.bsky.social! CJ is a member of the technical staff at Anthropic and former Stripe Developer Advocate. In his talk "Assign to Claude: Making an Agent a First‑Class Teammate in Your SaaS," he'll build one Rails app on Claude Managed Agents live. Come see him: shorturl.at/UtUm8 131
CJ Avilla @cjavilla.bsky.social · 18/08/2026Inspired by the CloudflareOS announcement I've been exploring the concept of self-edited software a ton. I've got a Rails app with a Claude Managed Agent that monitors prod/analyics, has MCP server to itself, can make PRs against the code where it lives. Blog post soon. 130
CJ Avilla @cjavilla.bsky.social · 17/08/2026Directory got a face lift to make it easier to connect Claude with all your favorite tools! 💅 claude.ai/directory 000
CJ Avilla @cjavilla.bsky.social · 17/08/2026Good talk explaining the evolution from Messages API to Claude Managed Agents piped.video/watch?v=K0X9QDRk… Link Evolution of agentic surfaces — Gagan Bhat & Isabella Kai He, Anthropic Sonnet 4.5 developed what Anthropic's Applied AI team came to call ... youtube.com 010
CJ Avilla @cjavilla.bsky.social · 15/08/2026🫰 tip: New doc out with a ton of advice about optimizing for cost vs intelligence! Check it out: platform.claude.com/docs/en/… Link Optimizing for cost and intelligence Balance cost and intelligence on the Claude Platform, with measured results... x.com/cjav_dev/status/2088637334472… 000
Reposted by CJ AvillaHildy Johnson @itshildy.bsky.social · 14/08/2026This Summer I had Claude build a very full-featured private iOS app for organizing and tracking one of my kid's collections, a sandbox video game for my other kid, and a couple GUI productivity utilities for me (one of which I opened sourced). Personal apps is 100% the future. 3101
CJ Avilla @cjavilla.bsky.social · 14/08/2026Great way to loop! Setup a loop to pull in TestFlight crashes, investigate -> repro -> write tests -> record demo of fix -> PR. 010
CJ Avilla @cjavilla.bsky.social · 13/08/2026Workspace IDs are in response headers, now! platform.claude.com/docs/en/manage-… 010
CJ Avilla @cjavilla.bsky.social · 13/08/2026🫰 tips: Learn a bunch of cost optimization tips in this cookbook. > Walk a claims-adjuster agent through a cost optimization checklist Prompt caching, input and output token management, agent loop tweaks, model and effort selection... x.com/cjav_dev/status/2087632786987… 000
CJ Avilla @cjavilla.bsky.social · 13/08/2026The dream of giving an agent a goal and a budget and letting it plug and chug is totally reasonable now. Checkout Claude Managed Agents session budgets + outcomes. Outcome = goal and rubric to see how it did Budget = set the $ you're down to pay... x.com/cjav_dev/status/2087594318269… 000
CJ Avilla @cjavilla.bsky.social · 13/08/2026So lucky I get to work with this team. Everyone is so fucking good at their jobs. It's hard to keep up! 010
CJ Avilla @cjavilla.bsky.social · 13/08/2026Don't sleep on the advisor pattern!! One roster entry, `{"type": "advisor", "model": "claude-fable-5"}`, and a Sonnet agent can pull Fable in mid-turn. consults are on their own thread, so you see when it fired and what it cost. Notebook:... x.com/cjav_dev/status/2086100130877… 000
CJ Avilla @cjavilla.bsky.social · 13/08/2026Sharing DevRel feedback used to feel like screaming into the void. Hear something on X, paste it into Slack, might as well have been a black hole. Now I paste the same thing and Tag often picks it up and runs with it the same day. 🥰 Feel so heard 000
Reposted by CJ AvillaLucian Ghinda @lucianghinda.com · 12/08/2026I have used the following two skills in multiple occasions in the last 6 months not only in personal projects but in professional enterprise ones and deploy them successfully: github.com/lucianghinda... and github.com/lucianghinda...github.comsuperpowers-ruby/skills/ruby-upgrade at main · lucianghinda/superpowers-rubyClaude Code superpowers: core skills library for Ruby and Ruby on Rails projects - lucianghinda/superpowers-ruby 111
Reposted by CJ AvillaSean van Zuidam @grimlink.com · 12/08/2026Many thanks to @anthropic.com for offering six months of Claude Max. It made building Fylgja UI a much easier process. 031
CJ Avilla @cjavilla.bsky.social · 06/05/2026RT Claude Re Effective today, we are: 1) Doubling Claude Code’s 5-hour rate limits for Pro, Max, and Team plans; 2) Removing the peak hours limit reduction on Claude Code for Pro and Max plans; and 3) Substantially raising our API rate limits for Opus models. 120
CJ Avilla @cjavilla.bsky.social · 05/05/2026My son thought of this and it was fun to build together: wordle for your ears. apps.apple.com/us/app/pitched-daily… 010
CJ Avilla @cjavilla.bsky.social · 24/04/2026RT Noah Zweben You can now /schedule one-time tasks from the CLI or the Routines UI via the Schedule -> Once Trigger. Use it to: * /schedule cleanup this feature flag in 1 week */schedule give me a launch report for my feature in 2 days 010
CJ Avilla @cjavilla.bsky.social · 23/04/2026RT ClaudeDevs Over the past month, some of you reported Claude Code's quality had slipped. We investigated, and published a post-mortem on the three issues we found. All are fixed in v2.1.116+ and we’ve reset usage limits for all subscribers. 020
CJ Avilla @cjavilla.bsky.social · 23/04/2026RT Noah Zweben You can now /schedule one-time tasks from the CLI or the Routines UI via the Schedule -> Once Trigger. Use it to: * /schedule cleanup this feature flag in 1 week */schedule give me a launch report for my feature in 2 days 000
CJ Avilla @cjavilla.bsky.social · 22/04/2026RT Katelyn Lesse if you’re building on the messages api, our explicit prompt caching is insanely powerful to improve latency and cost. even for the most sophisticated developers, there’s often more juice to squeeze. check out the new dash in... x.com/cjav_dev/status/2046906076084… 000
CJ Avilla @cjavilla.bsky.social · 16/04/2026RT ClaudeDevs For the developers building with Claude, a direct line from the team. Follow for changelogs, API releases, community updates, and deep dives. 000
CJ Avilla @cjavilla.bsky.social · 08/04/2026RT Claude Introducing Claude Managed Agents: everything you need to build and deploy agents at scale. It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days. Now in public beta on the Claude Platform. 110
CJ Avilla @cjavilla.bsky.social · 03/04/2026RT Lydia Hallie ✨ Re Digging into reports, most of the fastest burn came down to a few token-heavy patterns. Some tips: • Sonnet 4.6 is the better default on Pro. Opus burns roughly twice as fast. Switch at session start. • Lower the effort level... x.com/cjav_dev/status/2039844716720… 000
CJ Avilla @cjavilla.bsky.social · 19/03/2026😍 Right after authenticating with the Stripe CLI, they show you this little ditty: 020
CJ Avilla @cjavilla.bsky.social · 19/03/2026RT Jeff Weinstein Re Demo time! 🤖💸🎉 Here's @stevekaliski, @BackseatVC, and their agent (with a wallet) throwing together an actual birthday party. To accomplish this fun task, the agent autonomously researches, requests, and pays multiple businesses on Stripe via MPP. Note: No hands on keyboard 010
