CEO @CopilotKit🪁 | ex-@Meta | Author of PsyPhy (experimentally testable physics theory of consciousness) | Physics BA & MSc @Penn

Joined January 2019
243 Photos and videos
Pinned Tweet
We've raised $27M to build @CopilotKit — the Agentic Frontend Stack connecting humans & agents. Because all UI will be AI. Co-led by Glilot Capital, NfX and SignalFire.
55
52
409
325,537
PSA: Fable 5 does not listen to “keep subagents small” in prompts. runs that usually end in a few hundred sessions went to 44,796
Did the US gov shut down Fable because they too had en engineer accidentally burn through $22k in tokens in 12 hours?
3
499
Did the US gov shut down Fable because they too had en engineer accidentally burn through $22k in tokens in 12 hours?
1
4
706
Agents in @SlackHQ just got Generative UI PLUS streaming replies, tool calls, human in the loop, thread context and more How agents interact with users is changing. Soon, all UI will be AI
Slack 🤝 AG-UI Protocol @SlackHQ is now a first-class AG-UI frontend with: → Streaming replies → Tool calls → Human in the Loop approvals → Full thread context Same agent you built for the web, now living where your users actually work.
1
4
10
931
Your @SlackHQ agents just got a UI 🪁 via the AG-UI protocol. Early access below 👇
Slack 🤝 AG-UI Protocol @SlackHQ is now a first-class AG-UI frontend with: → Streaming replies → Tool calls → Human in the Loop approvals → Full thread context Same agent you built for the web, now living where your users actually work.
1
9
1,054
Daytona 🤝 CopilotKit 🎉 Excited to release our new cookbook In our first guide, your @CopilotKit Built-in Agent can run code inside a Daytona sandbox and return the result to your chat. Shoutout to @ivanburazin, @ivanda_mislav and the Daytona team for making this happen alongside Mark Fogle, @tyler_cpk and @sofiiiiiasz. Looking forward to more collaborations 🪁
3
8
23
1,587
Docs demo: docs.copilotkit.ai/cookbook/… Beyond running code, this opens up: → Reusing a sandbox → Running a server and previewing it → Custom environments, pre-loaded and ready → Any other language — Go, Rust, shell For example, you can ask your agent to build Snake and then play it in the chat. See below. What Daytona recipe should we do next?
1
6
122
Build your own Cowork/ChatGPT-like apps @CopilotKit open sourced Agentic Chat, Generative UI, and *now* Threads. Threads can capture generative UI, human-in-the-loop workflows, shared state, voice, files, and multimodal interactions. Cool demo: insight-canvas-rho.vercel.ap…
7
4
52
5,994
Built on AG-UI, so it works with any compatible agent framework like LangGraph, Google ADK, Microsoft Agent Framework, AWS Strands, Mastra, Pydantic, CrewAI and more. In one hook (useThreads) you get a synced thread list, rename/archive/delete, and realtime updates across tabs. 📄 Docs: docs.copilotkit.ai/threads ⭐ GitHub: github.com/CopilotKit/Copilo…
1
1
364
Where should @CopilotKit go next? Like Sofia said, many enterprise leaders from non-tech sectors (hospitality, healthcare etc) feel behind on implementing AI into their products. Our goal is to get these industries up to speed to what we see progressing in Silicon Valley. We'll go anywhere to make it happen.
that's a wrap on the conference portion of #LondonTechWeek! we spoke to people from 23 different countries, 59 different industries, and over 180 founders, engineers, and industry leaders at the @CopilotKit booth a common theme from many of the enterprise leaders from non-tech sectors (hospitality, healthcare, etc) is how behind they feel on implementing AI into their products. there is a huge market opportunity here to catch these industries up to speed to what we see progressing in silicon valley it was great to do this event alongside our marketing lead @eli_discovers! being a global company means we don’t always get face to face time with our fellow colleagues, so it’s always nice to have time to chat irl didn't get to meet us at this conference? we'll be at AI Engineer World's Fair in SF next month!
1
2
9
981
Atai Barkai retweeted
CopilotKit is now in Awesome LLM Apps (114K stars), in the new Generative UI and Agentic Frontends section. This covers where the agent renders, takes input, and streams state back to the user. That's what we build. Thanks @Saboo_Shubham_ for one of the best resources out there for agent builders. Check out the GitHub: github.com/Shubhamsaboo/awes…
3
8
57
3,390
Atai Barkai retweeted
One reason evals ended up not being a silver bullet at the app layer: AI output quality is largely subjective and personal. Legal contract review is more like a YouTube recommendation algorithm than something that can be “correct”. Imagine trying to build YouTube’s recommendation algorithm using evals: “If user watches Mr. Beast, then they should be recommended PewDiePie” “If user watches KPop videos they should see more KPop videos” “Every user should see Darude’s Sandstorm once a year because it’s a great song” Evals are the wrong tool to measure the quality huge chunk of AI outputs. Production analytics and A/B testing is much more useful a lot of the time, but no one talks about it because it doesn’t sound like real fancy AI work For example, in contract review the main thing we measure is: “what percent of the recommendations we make actually get accepted by the user?”
2024: finetuning is the moat 2025: evals are the moat 2026: model router is the moat Thin viral memes that give non-technical people a sense that a company is doing “real hard AI stuff” A rotating shell game—when one meme loses steam the next one comes in
17
2
98
17,521
🎉 @CopilotKit has been trending all weekend All fullstack agent apps in the future will have Agentic UI. Here are a few things people are getting wrong though: → Agentic UI isn't "regenerate every pixel with an LLM." It's an agent in the middle of every interaction. → Generative UI is a spectrum from more control to less: (MORE) invoke existing components -- assemble them -- embed apps via MCP -- generate UI from scratch (LESS) → Token cost is often less than a wall of text. Only the open-ended end is expensive. → The system learns — from user interactions and agent traces — to pick the right point on the spectrum over time. All UI will become AI, and our team is making that happen 👏
My job is to make @CopilotKit grow. I think we're doing that 📈👇 Today, our packages are getting MILLIONS of weekly installs. When I first joined, that number was around 50k. This kind of growth isn't trivial. With AI, building today is easier then ever. What HAS become *exponentially* more difficult in the last 2 years is being able to SIGNAL correctly from within all the NOISE. Our team works really hard to do that. Every week we're reaching new milestones. At this point, we're starting to get used to it... but it wasn't always like that. Since everything changes so fast today, you can't become complacent even for a day. We go into every week like it's a new battle. Now we are powering a level of usage that would've been hard to imagine one year ago. I'm excited for these next 6 months. If you're a dev who wants to easily give your app Agentic UI, CopilotKit is the best ecosystem tool. It's open source: github.com/CopilotKit/Copilo… Read more: copilotkit.ai/blog/frontend-…
3
6
25
2,072
Atai Barkai retweeted
All UI will become AI This is THE article you'll need give your agents a frontend. Agent-user collaboration is the future, see below 👇
44
106
1,390
496,226
It was a big week for Generative UI @CopilotKit just hit GitHub Trending #1 Everyone is realizing that agents will use UI to act and respond, not just walls of text. We're building the easiest way to put AI agents inside any app your users already live in (your web app, mobile app, Slack, Teams etc) and make them smarter with every interaction. See below.
All UI will become AI This is THE article you'll need give your agents a frontend. Agent-user collaboration is the future, see below 👇
6
11
87
12,206
Atai Barkai retweeted
What more is there to say? @CopilotKit #3 Overall Trending on GitHub 🔥
1
3
8
499
Atai Barkai retweeted
All UI will be AI. AI Agents need generative UI to EXPRESS not another paragraph of text.
19
36
286
78,396