AI agents are the same time brilliant and dumb.
Like, incredibly brilliant, and incredibly dumb.
It's hard to put it into words.
But this feels like alien technology.
I've created an agent skill for Tailwind CSS to teach your AI agent the v4 way:
- CSS-first config with @ theme
- New gradient syntax
- Opacity modifiers
Plus all other breaking changes so it stops generating v3 code.
Also includes tailwind-variants patterns and tw-animate-css.
AI can’t be left alone for a long time for the simple reason that we often don’t know what we want from a product until we start building it — and only then we discover where our initial guesses were wrong.
Sablier is one of the few web3 apps that correctly handles transactions that need more gas than the chain allows (block gas limit / EIP-7825).
We simulate with Tenderly, then show a beautiful error modal like this:
Our Sentry feed is getting noisy from errors thrown by web3 wallet extensions.
Is there an npm package/repo that maintains a curated list of "known wallet extension" ignore patterns (messages + stack traces) for filtering?
Excited to ship `@prb/effect-evm` - an Effect-TS library for EVM frontends.
Typed errors. Composable services. Real observability.
No more "transaction failed" with zero context.
Built on viem. Used daily at @sablier.
I love Claude Code, but they really botched it in v2.1.x
Tons of bugs and performance issues. And customer support is inexistent on GitHub.
I've switched to Codex for the time being.
Claude Code pro tip.
Use a snippets app on your machine to store XML-tagged prompt templates.
Then, learn a few keyboard shortcuts to instantly paste the templates in CC.
