Tony Sullivan

Always appreciate Nilay asking real questions The back and forth here related to integrating an LLM with Google Calendar instead of using an API was interesting.

I've seen more and more complaints over how aggressively web scrapers are hitting sites today Part of the challenge is them having to scrape rendered HTML.

I'm not sure where the #llm ends and the harness begins.

A month or two ago Claude failed hard when trying to read my own blog posts, I have Cloudflare's bot prevention enabled.

I'd really like a #js package manager that vendors everything.

🌶️ The #js ecosystem should never have moved to building and/or bundling shared packages Building, bundling, tree shaking, etc are consumer concerns, not publisher concerns Ship your source, only support #esm these days, and let the end user deal with the…

This billing change for Agent SDK and \`claude -p\` sure seems like anticompetitive behavior Starting June 15, I'll be able to use Anthropic's official Claude Code tool for coding on a subsidized subscription but for third party coding tools that otherwise

Years ago I wanted a #webdev framework that used state machines instead of state tightly coupled to UI components Well here goes nothing...

Started working on a #webdev last weekend #now ✅ Server rendering for all but transient client state ✅ State machines for all state management ✅ JSX-ish templates with read-only state access ✅ Events for all state changes ✅ Server send events + HTML parti…

Its very confusing that we keep hearing about this 60 day window in which Trump could effectively go to war without congressional approval.

I must be getting old and using outdated definitions, but I could have sworn that a ceasefire between militaries meant you no longer shot at each other.

I keep hearing in podcasts that NYC legally requires a balanced budget and/or that Mamdani inherited a $6B deficit.

🌶️ auto-detection for #signals dependencies in #js is a bad idea At runtime, you'll only discover dependencies in the execution flow. i.e. if your signal or effect logic is conditional in any way, dependencies may be missed.

I really want my #llm rules to amount to nothing more than "If there's a README.md read it.

I'm still not sure how we could claim alignment if we (a) don't know what happens during inference (2) can't define intelligence and (d) are limited to making polite requests via markdown.

It'll be really interesting to see how the budget fight in NYC plays out.

An even worse (maybe?)

🌶️ Spicy Friday Its been a good week so my spice level is low.

Anyone know of a well done prediction of what happens if we don't actually create AGI or ASI in the near future?

🌶️ XML + XSLT would make for such a better web experience, human or not ✅ schemas ✅ discoverable APIs ✅ data + optional templates for HTML