Atai Barkai

The Story

An agent that answers with a pie chart, not a paragraph

Ask Atai Barkai what is wrong with the chat box and he will not give you a manifesto. He will give you an example. An agent should be able to reply to you, he says, "not just with blocks of text, but with interactive UIs that are defined by your own company." Instead of a wall of words, a chart. Instead of a summary, something you can click. That small reframing is the whole bet behind CopilotKit, the Seattle company he co-founded and runs as CEO.

CopilotKit is open-source infrastructure for what Barkai calls agent-native applications - software where an AI agent lives inside the app, sees what you see, and acts in the same interface you already use. The center of gravity is AG-UI, an open protocol that standardizes how AI agents connect to and talk with user-facing applications. If MCP is how agents reach tools and A2A is how agents reach each other, AG-UI is the missing third leg: how an agent reaches a human.

That framing turned out to be unusually well-timed. In May 2026 the company announced a $27 million raise - a $20.5 million Series A on top of $6.5 million in earlier seed money - led by Glilot Capital Partners through its Glilot+ growth fund, with NFX and SignalFire alongside. By then AG-UI was not a niche library. It had been adopted by Google, Microsoft, Amazon and Oracle, and wired into AI frameworks including LangChain, Pydantic AI and LlamaIndex. The open project was clocking around four million downloads a week and roughly forty thousand GitHub stars.

Two brothers, one early bet

CopilotKit is a family operation. Atai is CEO; his brother Uli Barkai leads growth. The two are Israeli founders who incorporated the company - originally as Tawkit Inc. - and started building during the GPT-3.5 era, before "agents" was even the word everyone used. The premise was contrarian at the time and obvious in hindsight: humans and agents would end up building and running software together, so somebody needed to define the seam where they meet.

They publicly launched about two years before the Series A, shipped AG-UI roughly a year before it, and watched the rest of the industry walk toward the same conclusion. The line Barkai and his co-founder keep returning to is blunt: "all UI is becoming AI." They compare what is coming to what Cursor and Claude Code did for developers - except aimed at every vertical and workflow, not just the people who write code.

The unglamorous layer, on purpose

Barkai did not arrive at infrastructure by accident. He studied physics at the University of Pennsylvania, earning both a bachelor's and a master's there, with a stretch poking at the foundations of quantum mechanics and a summer of research with the university's High Energy Physics Group. Then he spent a decade building the parts of software that users never see and engineers cannot live without.

At Doximity he was a senior iOS engineer leading internal SDKs and the company's Dialer app. At Meta he ran a cross-team effort to build a media SDK, improving infrastructure that touched both Instagram and Facebook. Before CopilotKit he founded tawkitAI and built PodcastGPT, an AI podcast assistant that scaled to tens of thousands of audio hours streamed per month. The throughline is consistent: he keeps gravitating to the layer underneath the product, the standard rather than the app.

Why companies hand him their agents

CopilotKit's pitch to large enterprises is less about magic and more about freedom. "Maybe they're already using the Google, Amazon, Oracle, Microsoft, LangChain, Mastra stacks," Barkai says. "They want optionality, and they want self-hosting." AG-UI is positioned as open, independent and free - a neutral protocol rather than a walled garden - which is exactly why rivals as large as Google and Microsoft can all stand on it without anyone losing.

The customer roster reflects that. CopilotKit says more than half of the Fortune 500 use its tools, mostly through the open-source project but also as paying customers of its enterprise product, CopilotKit Enterprise Intelligence. Named users include DocuSign, S&P Global, Cisco, Deutsche Telekom and Function Health. The platform powers millions of agent-user interactions every week - the quiet kind of scale that does not trend but does compound.

What he is actually trying to win

Barkai's ambition is not to own the agent. It is to own the standard for how agents and people meet - and then keep that standard open so it spreads faster than any one company could push it. The strategic insight is that in a market where every cloud giant is shipping its own agent stack, the valuable, defensible position might be the neutral connective tissue between all of them. It is a patient bet, made by someone trained to think from first principles, that the interface layer is where the next decade of software actually gets decided.

There is a tell in how he describes the moment. The founders started, in his telling, with a single conviction held before the evidence arrived - that agents and humans would collaborate inside software rather than across a chat window - and then spent two years building toward a conclusion the rest of the industry only reached later. Shipping AG-UI a full year before the funding announcement, and watching Google and Microsoft and Amazon and Oracle each decide independently to build on the same open spec, is the kind of validation that is hard to manufacture and harder to fake. The downloads and the stars are downstream of the bet, not the other way around.

It also explains the discipline around keeping AG-UI free. A less patient founder would have tried to tax the connection - to make the protocol proprietary the moment it had leverage. Barkai's read is the opposite: the protocol is worth more to CopilotKit precisely because it belongs to everyone, and the company makes its money one layer up, where enterprises want self-hosting, support and the Enterprise Intelligence product rather than a closed standard. The open thing draws the crowd; the commercial thing serves the part of the crowd willing to pay. Holding those two ideas at once, without flinching, is the quiet skill underneath the whole company.