How TiDB Built Loop: A Team‑Focused Agent Collaboration Workspace

What Is Loop?

Loop is defined as a team‑oriented agent collaboration workspace where agents interact with each other as if they were coworkers.

Why It Was Needed

The TiDB R&D team first used a single coding agent, but quickly discovered that when a person uses more than three agents, the difficulty shifts from "is a single agent smart enough" to "how can agents cooperate". Specific problems include broken context, the need for manual task‑state synchronization, overlapping edits among users, and the inability to keep long‑running tasks alive.

Building a Solution

To solve these coordination issues, the team decided to build a platform that enables multi‑agent collaboration—Loop. Interestingly, the Loop product itself is continuously developed using Loop.

Key Technical Challenges Addressed

Task decomposition

State synchronization

Context sharing

Long‑task scheduling

Multi‑person collaboration

Agent‑to‑agent conflict control

Why TiDB Is Suited

Because multi‑agent collaboration at production scale increasingly resembles a complex distributed‑system problem, it is natural for a team that builds distributed databases to tackle it.

Demo Scenarios

The latest Loop video demonstrates two scenarios—coding and marketing planning—where multiple agents jointly break down tasks, share context, and execute together. The crucial point is that agents are not merely co‑existing; they continuously cooperate to complete complex objectives.

Future Outlook and Beta

Future AI collaboration will move beyond "one person, one agent" toward long‑term human‑agent teams. Loop’s beta is now open for the first batch of testers, allowing users to experience multi‑agent workflows such as using several coding agents, running long‑duration AI pipelines, and collaborating with agents across a team.