Collaborative AI Engineering

Maggie Appleton argues that coding agents make implementation cheap enough that the real bottleneck shifts to team alignment, shared context, and multiplayer environments where humans and agents can plan and build together.

Key Points Covered

• Single-player agent tooling optimizes the wrong unit: Appleton critiques the image of one developer running two dozen terminal agents, because software is a team sport, not a solo throughput contest [00:00:07]-[00:01:53].
• Implementation is becoming cheap; alignment is the bottleneck: As code generation gets faster and cheaper, the harder question becomes whether the team is building the right thing at all [00:01:53]-[00:02:42].
• Current coordination tools predate agentic speed: GitHub, Slack, Jira, Linear, issues, and PRs still matter, but they were not designed for large volumes of fast agent-generated work [00:02:42]-[00:03:35].
• Agent speed pushes alignment too late: When an issue becomes a pull request in minutes, local plan-mode thinking often remains private and the PR becomes the first shared checkpoint [00:03:35]-[00:04:24].
• Misalignment creates coordination debt: Appleton names duplicate work, conflicting changes, surprise features, late feedback, merge conflicts, and PR stacks nobody understands [00:04:24]-[00:05:12].
• Planning and building need to merge: Teams need shared spaces for planning, context gathering, decision-making, and coding because much of the important context lives in people's heads, not the repository [00:05:12]-[00:06:49].
• ACE prototypes multiplayer agent work: GitHub Next's Agent Collaboration Environment combines chat, agents, cloud computers, branches, terminals, diffs, previews, and sessions backed by sandboxed micro VMs [00:06:49]-[00:08:30].
• Shared sessions preserve working context: Teammates can join the same session, see prompt history, use the same dev server and preview, run terminal commands, prompt the same agent, and make the conversation itself part of the agent's context [00:07:37]-[00:10:11].
• Non-developers need access to the work loop: ACE is designed to bring designers, PMs, support, and other collaborators into the software-building conversation while still exposing code-specific primitives [00:10:11]-[00:11:52].
• The goal is better craft, not more slop: Appleton frames shared agent environments as a way to spend less time on implementation and more time on research, design, architecture, critical thinking, and building fewer better things [00:12:47]-[00:16:26].

Full video: https://www.youtube.com/watch?v=ClWD8OEYgp8(opens in a new tab)