By CopilotKit just raised $27 million to kill the chatbot era. The Seattle startup’s thesis: AI agents shouldn’t live in separate chat windows. They should embed directly inside applications, understanding context and taking action like human users would. On May 5, the company closed a Series A led by Glilot Capital, NFX, and SignalFire. Fortune 500 companies are already deploying this in production – Deutsche Telekom, DocuSign, Cisco, and S&P Global. The open-source AG-UI protocol behind CopilotKit is seeing millions of installs per week. This isn’t a pilot. This is the infrastructure shift from conversational sidebars to embedded agents that actually do things.
The Chatbot Problem
Current AI chatbots sit in separate windows. You ask questions. They answer with text. They don’t understand what you’re doing in the application. They can’t execute workflows or interact with app state. Every action requires human intervention.
App-native agents change this. They see your application state. They understand context. They take autonomous actions within the interface itself. No back-and-forth. No copying and pasting between windows. The agent operates inside the app like a user would.
The data backs this up. Gartner reports 80% of enterprise applications shipped or updated in Q1 2026 now embed at least one AI agent, up from 33% in 2024. But there’s a gap: only 31% of enterprises have agents in production. That’s the infrastructure problem CopilotKit solves.
AG-UI: The Open Protocol
CopilotKit’s technical foundation is AG-UI, an open protocol for connecting AI agents to any user interface. Think of it as TCP/IP for the agent-to-user layer.
The architecture is straightforward. It uses HTTP and Server-Sent Events (SSE) – plain web tech, no special protocols needed. The front-end sends a single HTTP POST request with the user’s prompt or current state. The agent streams back standardized event types as it works. The entire exchange runs through standard infrastructure: firewalls, proxies, CDNs. No proprietary transport layer.
What makes AG-UI powerful is backend flexibility. You can swap GPT-4 for LLaMA without touching the front-end. Change your agent framework from LangChain to Microsoft’s offering. Switch cloud providers. The UI doesn’t care. The protocol abstracts all of that away.
This matters because enterprise deployments need options. Vendor lock-in is a deal-breaker when you’re building critical workflows. CopilotKit’s co-founders, Atai and Uli Barkai, designed the system to support “whatever agent framework, cloud provider, or back end an enterprise already uses.”
AG-UI sits in the broader agent stack alongside Model Context Protocol (MCP), which handles agent-to-tool interactions. MCP hit 97 million downloads and became the de facto standard for tool connectivity. AG-UI aims to do the same for the human-in-the-loop layer.
Fortune 500 Validation
This isn’t a prototype. Fortune 500 companies are running CopilotKit in production right now.
Deutsche Telekom is deploying it for telecommunications workflows. DocuSign is using it to “enable in-application agentic experiences that improve workflows and clarify customer questions.” Cisco has it in networking and collaboration tools. S&P Global is running it for financial data operations.
The company reports millions of installs per week and says the majority of Fortune 500 and Global 50 companies are using the protocol in production. That’s not pilot territory. That’s enterprise adoption.
This validation de-risks the technology for other developers. If Deutsche Telekom and DocuSign trust it in production environments, it’s production-grade.
What $27M Enables
The funding launches CopilotKit Enterprise Intelligence, a self-hostable product that packages the infrastructure businesses need to deploy agents at scale.
Enterprises get persistent conversation threads, analytics, observability, real-time learning capabilities, and the governance and compliance tooling that procurement teams require. The self-hostable architecture keeps data on-premises for organizations with strict security requirements.
The business model is clean. The AG-UI protocol remains open-source and free. The commercial product hardens the stack for enterprise needs without gating the core technology. This isn’t an “open core” trap where production features get locked behind paywalls. The protocol stays open. The commercial layer adds enterprise operationalization.
CopilotKit positions itself as infrastructure, not a full-stack platform. It supports any agent framework – LangChain, Autogen, Microsoft Agent Framework, whatever you’re already using. No vendor lock-in on models or cloud providers.
The Market Shift
The AI stack is maturing from “answer questions” to “take action.” Chatbots were the assistant era. Embedded agents are the automation era.
The AI agent market hit $10.9 billion in 2026, up from $7.6 billion in 2025, growing at 45%+ CAGR. Median time-to-value is 5.1 months. Banking and insurance lead adoption with 47% of companies running agents in production. Healthcare and government trail at 18% and 14% respectively.
The gap between embedding agents (80%) and production deployment (31%) represents the infrastructure opportunity. Research shows only 12% of agent deployments succeed. Those who do succeed share one characteristic: they invest in infrastructure before deployment. Governance, observability, and baseline metrics are critical. CopilotKit provides that missing layer.
Embedded agents are becoming a standard feature of enterprise applications, just like databases and APIs are today. CopilotKit’s open protocol democratizes that deployment.
