Agents at Scale: The Manus Story and What Comes Next for AI

Description

Most agentic AI demos impress; too few survive real traffic. In this TiDB SCaiLE 2025 keynote, Manus VP of Engineering Ziming Miao traces Manus’ journey from ambitious prototype to a general AI platform built for real-world scalability. He shows how Manus’ agentic AI system plans, executes, and verifies complex tasks inside its own computing environment, then scales that agentic AI across massive demand. From launch to multi-million waitlists and rapid revenue growth, Ziming walks through the architecture, the bottlenecks, and the fixes that turned a viral prototype into a durable platform.

Discover how Manus activates latent model intelligence with context engineering loops (planning → execution → verification), why “more tokens” only helps up to a point, and how teaching the agent to treat the file system as long-term context preserves plans, actions, and outcomes (even across tasks that run thousands of iterations). Ziming details the storage explosion that followed paid growth and the company’s decision to migrate the entire data system to TiDB in two weeks with minimal code changes.

Ziming discusses how the strategic partnership with TiDB underpins Manus’ evolution. By aligning on a unified data foundation, Manus enables persistent memory and real-time insight without sacrificing oversight. Ziming highlights what’s newly possible with Manus today and outlines a forward view of agentic AI: one where agents plan, remember, and act with accountability, reliability, and measurable impact.

You’ll learn:

The building blocks of a production agentic AI system
How Manus uses caching, files, and retrieval to keep agent memory coherent
Hard-won infra lessons from a full data-platform migration- A look at “compute-as-conversation” and agent swarms for complex workflows