been thinking about this since last night a problem with appealing to a big audience used to be that you'd come up with a single system that's broadly applicable, so you end up with lowest common denom of solution, making it hard to build specialised experiences fuck it, build a thousand specialised platforms/solutions instead, isn't that what acceleration is meant to enable not saying it'd be perfect, but you'd end up with a better class of problems than the status quo

discovered 2 new usecases for subagents that have changed my focus for the next couple of days I really do feel like we have the best subagents in the world aim to have one demo by EOD, wish me luck

Cloudflare's security team spent the last few weeks testing Anthropic's Mythos against fifty of our own repositories. What we learned about offensive AI, why faster patching is the wrong reaction, and what the architecture around vulnerabilities has to look like next.

some things I'm working on this week on agents: - reviewing @mattzcarey's PR for postgres backed sessions (so you can use Think with planetscale etc) - reviewing @whoiskatrin's PR for reusable browser sessions in an agent - hardening subagents prepping for the new starter kit (multi chat/agent tools/etc) - polishing up the "skills" story (should we just let you import/use them freely?) - working on filed bugs, those are usually straightforward - thinking about devx (should we build a nextjs/tstart type framework on top of Think?) should be a chonky release, let's aim for friday.

big +1. conceptually this removes a lot of questions around preview environments, staging urls, testing with "real" data, version skew, etc. you gotta be a little careful with database schemas and migrations, but you should already be doing that tbh.

Is anyone doing feature flag development with agents? Not tried it, but in theory feature flagging is an alternative model to PR's to getting work on main. 1. Put it on main, disabled by a flag 2. Deploy with the rest of the system 3. Unflag to selected users early 4. Fix bugs for those users 5. Unflag to more users 6. Repeat until shipped Feels like a perfect strategy to pair with agents

imo the intersting thing about the bun rewrite (which is dope btw, and you ragers have been reacting embarrassingly), is the possibility of a parallel universe where the node.js team gets a million dollars in tokens and a full time dedicated team

tl;dr - subagent behaviour working on adding multi chat and subagents to the agents starter (yay!) and I have a curious product direction/question. our subagents can be full fledged chats themselves. which means they could not only be async while they work on their thing and you continue, but you could continue "talking" with them after they've "returned" a result. so what should the default behaviour in the starter be? - readonly, no input. this is what most (all?) products/devtools like this do atm - have chat, but it's only followups, doesn't affect the main chat - add a "send back/summarize to main chat" this feels powerful and underexplored I'll probably ship option 1 for now, but there's something here... anyway, multichat/subagents in starter template coming this week