Rahul Dobhal

❤️ Big fans of @hwchase17 and the entire team at @LangChain. Cannot be happier to see this integration go live within a week of OpenUI release. Thanks a lot @bromann for putting this together

I am very bullish on OpenUI (openui.com) .. it comes with great components out of the box, streaming protocol is super efficient and has a super small footprint! Also .. very easy to put together with @LangChain_JS 🚀

Generative UI agents on iOS and Android @reactnative support is live in OpenUI You shouldn't need a separate stack to bring Generative UI to mobile.

getting rid of your smartwatch is the first step towards internal peace

We deleted entire Rust codebase and replaced it with TypeScript and it got 3x faster. Rust is fast != Rust in browser is fast Here's what we found 🧵

watch openai acquire them in like 6mo

We just open sourced the core rendering engine behind C1, our generative UI engine. The result is: • 3x faster rendering • 67% fewer tokens • Works even with small models running locally

retweeting to save, will try later

posting for the odds of getting lucky

3. pulling product analytics and acting on them without figuring out dashboards (saves ~1hr every time) lmk if you want templates for these. so that it takes ~4 clicks to set up.

launched agent builder by Thesys. kind of wild but building agents now feels faster than saying “let me get back to you”. and it responds with charts and cards instead of long text. things i’m personally using it for 👇

2. checking updates across google sheets while staying out of spreadsheets as much as possible (helps my eyes)

1. catching up with “can you do this?” messages into actual tasks in linear (saves my sanity)

When majnu kissed feet of a dog, people asked “What on earth are you doing?” Majnu said: “I’ve seen a similar dog pass by layla’s street ” Followed by depiction of insanity of Majnu Video Credit @dreamjourneyflm

MCP Apps feels like a new primitive for agents. We built @thesysdev MCP so Generative UI becomes a first-class citizen in agent workflows, not an afterthought. When your thinking happens inside @claudeai, it shouldn’t stay trapped in text. Humans reason spatially, visually, iteratively - agents should meet us there, generating charts, cards, slides, and reports as part of the thought process itself. Built on MCP Apps. Brought to life inside @AnthropicAI’s ecosystem. This is how agents become interfaces.

Work is changing. It’s no longer tickets, dashboards, and slide decks stitched across disconnected tools. MCP redefined the data layer. With MCP Apps, @claudeai and @anthropic are redefining the application layer. Integrating @thesysdev was the next obvious step. This is the Generative UI layer for a modern Work OS.

wha

Great to see the world moving more in this direction. When we created C1, most people asked us: why can’t I just generate code and render it? Direct codegen without a catalog creates more problems than it solves. How do you maintain consistency, responsiveness, handle streaming, interactivity, and more without a formal spec for the LLM to reason against? Without that spec, the model is guessing. The result is inconsistent UI, hallucinated actions, and fragile states that break as soon as you introduce streaming or partial updates. That’s why we created C1 DSL: an intermediary spec that allows LLMs to generate consistent, performant, interactive outputs by construction. As the ecosystem starts to think of Generative UI as core infrastructure and not just a UI layer slapped onto chatbots, direct codegen will start to look naive. 2026 might just be the year of Generative UI after all.

gpt and me