The joke in the video about needing 7 YC startups to do one agent task is actually a good description of the current stack.

Xpander's model is simpler: publish an agent once, then let authorized coworkers invoke it under their own permissions.

Here, you can hand Omni a job in plain language and have it build the agent, but if you want, you can still build your own agents in code with LangChain, Strands, Agno, or another framework.

Either way, the resulting agent lands inside the same runtime for memory, deployment, permissions, observability and tool access.