Making AI multiplayer with Maggie Appleton from GitHub Next @ AIE Europe
Maggie Appleton argues that faster solo coding creates an alignment crisis: the next generation of agentic development environments must make planning, context and execution multiplayer from the beginning.
Maggie Appleton works at GitHub Next, GitHub's research-and-development lab. At AI Engineer Europe she presented ACE, an Agentic Collaboration Environment designed around teams working with coding agents in shared, real-time cloud workspaces.
We discuss why pull requests are too late for alignment, how shared sessions expose the path from conversation to code, what teams learn by watching one another prompt, and the possibility of proactive agents that connect developers with the right organizational knowledge.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What is wrong with today’s single-player coding agents?
Giving one developer two dozen agents can multiply that person's output without aligning the team. People can build duplicate features, edit the same files or ship work nobody requested. Software depends on communication and coordination, so accelerating isolated execution can produce a larger volume of mediocre, conflicting work.
Why must collaboration happen throughout development?
Planning and building form a continuous loop: make a prototype, discuss it, learn why it is wrong and try again. The deciding context includes leadership, business constraints, design, user research and customer knowledge that does not live in the repository. A private agent plan cannot capture what a whole organization knows.
Why can’t pull requests solve the alignment problem?
A pull request appears at the end, after the key decisions and much of the work are complete. Reviewers see code or perhaps a generated markdown plan, but not the conversation that produced it or whether an idea came from the developer or the model. That makes feedback expensive precisely when changing direction is hardest.
What is ACE?
ACE is a GitHub Next research prototype that feels like Slack with coding agents. Teammates and agents share sessions, chats, cursors and plans. Everyone can prompt together, and every session connects to a cloud microVM where the team can run commands, start the development server and see the same results.
Why put the development environment in the cloud?
Local branches and worktrees make it awkward to invite a coworker into the exact state of a task. A shared cloud computer removes the stashing, checkout and explanation ceremony. Parallel streams remain separate, but another person can enter the live environment, inspect the same output and contribute immediately.
Does more organizational context make the agent code better?
Not necessarily in a narrow capability sense; coding agents are already able to implement a great deal. The improvement is choosing the right work. A feature should reflect what users need, what product leaders intend, what designers understand and what the business can sustain — not merely what a repository and one prompt suggest.
How does a multiplayer environment change team behavior?
The lowest-friction action becomes asking a colleague for an opinion inside the session. Teammates see one another's prompts, plans and techniques, creating useful knowledge sharing. Open-by-default work needs private spaces too, particularly when somebody is learning or does not want a senior colleague hovering over every question.
What can proactive agents add beyond shared coding?
A dashboard can restore context each morning, summarize what teammates changed and warn when two people are approaching the same area. The larger opportunity is to connect work with conversations elsewhere — bringing in a researcher, salesperson or domain expert at the useful moment. The hard product problem is doing that accurately enough that people do not mute the agent as noise.