AI in Action: The Ultimate Live Demo with Theia AI

What can go wrong with a 40-minute, all-live, all-AI demo session at a conference, running against a remote LLM, with the network in the room, in front of a packed audience? Quite a lot, in theory. At Open Community Experience 2026 (OCX26) in Brussels, we opened the OC for AI track with exactly that kind of adventurous session: no slides, only live demos with Theia AI and the Theia IDE.

Spoiler: almost everything worked. The full recording is now available:

What the Session Was About

The talk was a guided tour of what’s possible today with AI-native development tools as a sequence of real, end-to-end demos. The promise of “no slides” was kept by a small twist: the few slides that did appear were generated live by Theia Coder from a markdown context file.

Across the session, three pieces of technology were on stage:

• Eclipse Theia platform: the open framework for building IDEs and domain-specific tools.
• Theia AI: the part of Theia for building AI-native tools (agents, tools, MCP, capabilities, sub-agents, skills, chat UI).
• Theia IDE: the open, AI-native end-user IDE built on top, used as the demo stage.

Everything you see in the Theia IDE, you can build similarly into your own domain-specific tool.

Key Themes from the Demos

Here are the ideas worth taking away:

• Capabilities: one-click toggles in the chat that bundle MCP servers, sub-agents, and prompt fragments into high-level abilities (E2E testing, GitHub integration, shell access). As a tool builder, you decide what your users see.
• Sub-agents: self-contained tasks like “test this app” or “report this issue on GitHub” are delegated to dedicated agents with narrow context and tools. Less context pollution, more reliable results.
• Skills and slash commands: focused knowledge or workflows you can load into an agent on demand. In the demo: turning a fresh app into a retro Eclipse-styled UI, or fixing a GitHub ticket via /fix-github-ticket.
• Full customisation: every built-in agent is open. Read its system prompt, edit it, or create your own agent in seconds via the configuration view.
• Plug in anything: external agents like Claude Code can be integrated into the same UI, with custom chat widgets rendering their interactions natively.
• AI-assisted code review with the developer in the loop: as an example for a custom workflow, a PR Reviewer agent that walks the developer through findings file by file, with confirm/skip buttons, and posts a clean review back to GitHub. The human keeps the judgment; the AI handles the mechanics.
• Beyond coding: the same building blocks drive a diagram editor. An AUTOSAR-style function block diagram is generated from a textual system description, then queried and modified through natural language.

If you want to see how all of this actually feels in practice, including the small live mishap that got fixed on stage, the video is the better medium for it than this blog post.

Want to Build Something Like This?

If your team is thinking about adding AI capabilities to a tool or IDE (for coding, modelling, configuration, or any other domain), we’d love to help. We do this for a living:

👉 AI-Native Tool and IDE Development: design and implementation of custom AI agents, capabilities, and workflows on top of Theia AI.

👉 Tool and IDE Development based on Eclipse Theia: building custom IDEs and domain-specific tools, with or without AI.

👉 Domain-Specific AI Solutions: integrating AI deeply into business applications and workflows.

👉 AI Coding Training: help your team adopt AI-assisted development with a systematic, proven methodology.

👉 Contact us to discuss your use case.

💼 Follow us: EclipseSource on LinkedIn

🎥 Subscribe to our YouTube channel: EclipseSource on YouTube