Source note

One Developer, Two Dozen Agents, Zero Alignment

Multi Agent Software EngineeringCode IntelligenceCollaborative AIDeveloper ToolsHuman AI Interaction

Ace is a GitHub Next research prototype for multiplayer coding with shared agents, chat, planning, and cloud workspaces. The talk argues that the main bottleneck in agent-driven software work is team alignment, not code generation speed.

  • Current coding agents are built as single-user tools, but software teams need shared planning, context, and coordination.
  • Faster code generation shifts the bottleneck toward deciding what to build, avoiding duplicate work, and keeping teammates aligned before code lands in pull requests.
  • Existing tools such as PRs, issues, and Slack do not handle the speed, volume, and shared context needs of agent-heavy development well.
  • Ace puts humans and coding agents in the same realtime session, with chat, prompt history, summaries, and shared context visible to everyone in that session.
  • Each session runs on a cloud microVM with its own git branch, terminal, dev server, preview, commits, and diffs, so multiple people and agents can work on the same task without local setup friction.
  • Teammates can join a session, inspect the full agent conversation, run commands, view the same preview, edit code together, and co-prompt the agent.
  • For larger tasks, the team can co-edit a plan inside the workspace before asking the agent to implement it, moving alignment earlier in the workflow.
  • Ace also adds dashboard summaries such as unfinished work, teammate activity, and recent repo changes to help users stay oriented across many parallel agent-driven tasks.
  • The excerpt provides no formal benchmark results, dataset evaluations, or controlled quantitative comparisons.
  • The strongest concrete claim is product-level: Ace supports one developer working with "two dozen agents" in a shared multiplayer workspace.
  • The prototype is described as entering technical preview with planned user testing by a few thousand people.
  • Claimed user-visible effects include shared cloud execution, multiplayer prompting, collaborative plan editing, realtime previews, automatic commits, GitHub PR creation, and persistent sessions that continue after one user disconnects.