Ben Weis

Swapped tsc for effect-tsrs on Springbird's monorepo typecheck. turbo run typecheck, 210 tasks, no cache: ⏱ 88s → ~45s (−48%) 💻 CPU time −46% 🧠 Peak memory 40 GB → 24 GB (−40%) Same checks, half the wait. Huge thanks to @EffectTS_ and @yielded_dev

This falls out of @EffectTS_ for free. The proof is a service in the requirements channel, the check is a Layer that fails with a typed error, and the route doesn't compile until every proof is provided. Same Vercel password example, side by side, no extra dependency 👇

Claude just had a “no way that just compiled on the first go” moment

So the actual js executes within a createNodeIsolateDriver from tanstack/ai-isolate-node. You can then further force the restriction of the vocabulary available by injecting a prelude restricting globals/intrinsics. See this quick ai prelude example.

Awesome! Took a crack at rebuilding this on Temporal, without the DSL. The model writes plain JavaScript. It runs in a sandbox with five functions available to it, and the script is rerun from the top every time one of those calls finishes. Calls that already happened return their saved result; the first one that hasn't is where the run waits. All of it is one workflow on effect-temporal.com/. Sleep is a Temporal timer, approval waits on a signal, a tool call is an activity. Still a prototype, but plain JS has something going for it.

“The Year of Effect So Far™️”

“What is a workflow, if not a job persevering?”

You can now run plain @EffectTS_ workflows on @temporalio: effect-temporal 0.2.0 hosts Workflow.toLayer registrations behind one workflowBundle default export. same workflow code runs on effect cluster or Temporal, you just swap a layer. thanks @MattiaManzati for the nudge, you were right effect-temporal.com