Taking it a step further you wouldn't need Agents\.md or .cursor/rules, you could manage your coding agents the same way you write the rest of your code. The agents could also be instructed to update their own flows, and you'd retain a deterministic view of the data flow.

What if you could use the same reactive data flow primitives for server, client, and even AI orchestration?.

Here it is:.

It only takes 160 lines to build a system supporting a node tree with reactive hooks.

Beauty of Balanced Ternary

Reactive dependencies can be explicit without being static. You don't need hidden global state to track what reactive values link together on the graph.

Ember's approach wasn't perfect. Reactivity was restricted to object properties, like React hooks dependencies were static and declared in both dependency list and logic. Solving these issues doesn't require throwing out visibility of the graph.

It's quite frustrating and I think part of a larger web tooling trend towards "it just works" as opposed to "it works and it's pretty clear how".

When I look at reactivity what I see is the graph. It's been that way for me since my early days with ember when reactive dependencies were explicit. Today most reactive libraries (ember included) make this very hard/impossible since they treat dependencies as implicit.

LLMs really do like to make shit up.

Building a efficient DOM renderer on top of such a system is comparatively easy.

Designing a reactivity system that supports dynamic dependencies, direct non-tearing granular projections, granular collection derivation, optional static dependency optimization, and async — is very hard.

Non-exhaustive wishlist for a TypeScript successor:. - Compatibility with .d.ts for easy adoption.- Removal of `any` and `void`.- Distinction between:. - Own properties and prototypical properties. - Value properties and accessors at a type level. - Pure/impure functions.- Use.

You can probably think of a time when you saw or participated in a poll with biased questioning. Remember that when you prompt a text generation machine, they are much more susceptible to the same effect. Too many people posting "LLM_NAME agrees with STATEMENT".

TSC shouldn't have been ported to Go. It should have been rewritten in TS.

Why do (AI) CEOs always have to ruin things with their sketchy marketing tactics. LLMs are very powerful cool tech, you don't have to go spouting crazy false claims to get people excited.

Magic is maybe an overloaded term in programming. What @DHH is describing I'd consider an intuitive abstraction. Something that whether by name and/or with a brief description gives you a concrete map of how it works. When I hear magic though, I think of hidden behaviours and.

LLMs or even state of the art AI agents cannot distinguish fact from fiction when it comes to niche technical topics. In general humans can't either. So what's the answer then?. I'm not entirely sure, but I think it'll come from looking at what predates search. Maybe it's just.

Is the web slowly breaking under pressure? . I think that's the case for the globally searchable web. There's overwhelming amounts of content, and the old methods of search ranking are failing, especially for narrow search queries. How can it be fixed? AI search? Probably not.

I knew I forgot to label something. A red node means it contains an expensive computation.