Learning Center
Plain-language explanations of the concepts behind agentic applications.
Agentic applications
What agentic software is, and how it differs from the software that came before it.
- The three buckets of the agentic worldAgentic software splits into three buckets: agent orchestration, agent-user connectivity, and agent learning. A mental model for locating any tool in the stack.
- What is agentic UI?Agentic UI is interface built for humans and agents working in one application: streamed agent work, mid-run steering, shared state, and generative UI.
- Agent-native applicationsAgent-native applications treat agents as first-class actors: shared state, real actions, generative UI, every surface, and a learning loop.
- Copilots vs. agentsAgents pursue goals autonomously; copilots are agents embedded in an application, working alongside the user. The difference is the surface, not the tech.
- Why agents break request-responseAgents run for minutes, stream output, take input mid-run, and outlive dropped connections. Why request-response fails them, and what replaces it.
Protocols
The open protocols that wire agents to their tools, to each other, and to the people they work for.
Interaction patterns
How humans supervise, correct, and collaborate with agents while they run.
- Human-in-the-loopHuman-in-the-loop lets an AI agent pause mid-run for human review, approval, or correction. The interrupt patterns and what the interface must do.
- What is agent steering?Agent steering lets users redirect a running AI agent mid-run: new instructions merge into live work, no restart. What it is and what it requires.
Agents in production
What happens after launch: learning, improvement, and accountability.
