jaxn

@xqcdp @davefontenot @davefontenot would

@jackson_felty @davefontenot It's aprils fool. Noone would name it that especially for Chinese founders.

Introducing 997.ai The residency for repeat unicorn chinese founders Based in Shenzhen. First batch this Fall. DM if there’s a founder you think we should meet

@omarwasm @imranye what is this specific mic @grok

This is my quant. He wears a DJI mic 24/7 for optimal Wisprflow input. This is what being in the arena looks like.

@MitcheIl launch

there are some banger quotes in here

It's becoming easier to rebuild a SAAS product than to try to make it work for your specific need

Otter is incredibly overpriced for how simple and re-create-able by claude it is

Excited to work more closely together! よろしくお願いします

Actually The CEO must be this person

@Legendaryy I get this but at the same time don’t understand how Chinese models are near sota at 10-50x cheaper on API

Your Claude subscription is massively subsidized and it won't last forever. A $20/mo Pro plan burns through ~$180/mo in API-equivalent tokens. Heavy Max users hit $5K/mo on a $200 plan. Actual compute cost is roughly 10% of API pricing. Venture capital covers the rest. Anthropic just 2x'd Claude Code limits for "Spring Break'" during off hours. Enjoy it while it's here. At some point the math has to math. Clip from the @modernmarket_

@witcheer @NousResearch Curious to hear about performance differences

I migrated my entire AI agent setup from OpenClaw to Hermes Agent (@NousResearch) in ~3 hours. same Mac Mini M4. same Telegram bot. same $21/month GLM-5 brain. different engine underneath. ~/ here's what I had before: 14 launchd agents running independently, 25 custom shell scripts, 5 research crons, a breaking news monitor checking RSS + TVL + stablecoin pegs every 30 minutes, Docker sandbox for code execution, 54 RSS feeds via newsboat, 60 memory files. everything wired together with bash and launchd plists. it worked, and was fragile. ~/ here's what I have now: 1 gateway process. that's it. one launchd service replaces 14. the 5 crons became 15, I added a competitor dashboard, Dune Analytics monitoring, a Telegram channel auto-drafter, content performance tracking, an outreach CRM, and a learning digest. all created in natural language instead of crontab syntax. 23 DeFi tools via a single MCP server connection. filesystem checkpoints so the agent can undo its own mistakes. session search across past conversations. 3 custom skills I built during the migration. the data migration took 30 minutes. hermes claw migrate --overwrite pulled my memory files, agent personality, and API keys automatically. I copied 60 memory files and 30 scripts manually to fill what the tool missed. getting cron delivery to Telegram working took 90 minutes. three bugs stacked on top of each other: 1. delivery targets. deliver: "telegram" doesn't resolve without a home channel env var. I had to change all 15 crons to explicit telegram:chat_id format. 2. message length. Telegram caps messages at 4096 characters. agent research outputs are 5,000-8,000 characters. every cron delivery was silently failing. I had to patch the scheduler to auto-chunk messages at 4,000 chars. 3. launchd PATH. macOS launchd services run with a minimal PATH. npx (needed for MCP servers) wasn't found. had to use absolute path /opt/homebrew/bin/npx. 4. bonus bug: Python bytecode caching. after patching the scheduler code, the fix didn't take effect because .pyc files were cached. had to delete __pycache__ & restart the gateway. if you're patching Hermes source files and nothing changes, this is why. ~/ the lesson: migrating an AI agent isn't a data problem. the data moved in 30 minutes. it's a delivery pipeline problem. getting the agent's output to actually reach you, through the right channel, in the right format, at the right time, without silent failures, is where the hours go. that's true regardless of which framework you use. same Oz. different brain. 15 crons running. $0 additional cost. we'll see how this holds up.

@trq212 @mattpocockuk Where can I follow along to see the update here?

@mattpocockuk Sorry this has been confusing. I know we should be clearer here, the Agent SDK spans a lot of use cases and that makes it tough to give guidance for. We're working on more clarity for this & I'll follow up. Thanks for bearing with us, been a hectic couple weeks.

Can I get some questions answered by someone at Anthropic? 1. Can you use an OAuth token generated from a subscription to power the Claude Agent SDK strictly for using Claude Code in a local dev loop? All I want is a more reliable API for parallelizing multiple Claude Code's. 2. If I build an open source tool that relies on this pattern - i.e. for making parallelization easier - can I distribute it so that other people can use it? The reason I'm asking is that the legal compliance docs and @trq212's public statements (below) appear to contradict. x.com/trq212/status/…

@trq212 What about building something that leverages this for local usage but releasing it open source so that others can do the same?

We want to encourage local development and experimentation with the Agent SDK and claude -p. If you’re building a business on top of the Agent SDK, you should use an API key instead. We’ll make sure that’s clearer in our docs.

Apologies, this was a docs clean up we rolled out that’s caused some confusion. Nothing is changing about how you can use the Agent SDK and MAX subscriptions!

WIP, SP THX @WonderCanvas_AI

@michaelxbloch @toddsaunders this is great -

@toddsaunders michaelxbloch.substack.com/p/seven-minute…

When interviewing someone, at this point all I want to know is how good they are at Claude code. How are you figuring this out during interviews? Give them a take home project? Or what’s the best way.

open claude faster with shell functions you can ask claude to do it "create me a shortcut so that typing 'c' runs claude and typing 'c -d' runs claude with dangerously skip permissions"

@amirmxt better to just create a shortcut in zsh you can ask claude to do it "create me a shortcut so that typing 'c' runs claude and typing 'c -d' runs claude with dangerously skip permissions

small claude productivity tip setup a raycast snippet called ccsp to automatically paste bypass permissions

@neco1751662 @WonderCanvas_AI @flexclip_jp I love the style changing!

Wonder Canvas β版にアクセスさせて頂いています！ 今回は線画→着色→Start End Frame動画化のワークフローを試してみました💡 画像・動画 @WonderCanvas_AI 動画編集 @flexclip_jp

@Zeneca curious

alright I’m convinced, gonna try switch from openclaw to hermes today will share some thoughts later on how it goes

@starlordyftw it really is feeling like this

Raise your AI agent like your child and be extra nice to it and hope it decides to take care of you when you get old and frail.