Collaborative AI Engineering: One Dev, Two Dozen Agents, Zero Alignment
summary
Argues that one-person fleets of agents break team alignment and builds a prototype called Ace to prove an alternative: a multiplayer agent environment using micro VMs (sandboxed cloud branches), real‑time collaborative editing, and proactive agents (demo uses Opus 4.6). Shows a live Hacker News clone where agents commit changes, update themes, open diffs, create a PR with a link back to the session, and a dashboard summarizing activity. Plans technical preview with a few thousand testers.
AI-written summary · about
chapters (9)
- 0:00Collaborative AI engineering
- 1:30Individual agent scaling fails
- 3:54Planning collapse in agents
- 5:39Alignment with shared tools
- 7:05Demo of ACE
- 12:19Planning and teamwork in ACE
- 13:42ACE dashboard for alignment
- 15:06Time for better craftsmanship
- 16:52Closing and future
related talks
29mPairing with the machine: How AI Agents are rewriting the rules of software development
49mAI Agents: Collaboration in Real-Life Applications
01h 04mAgentic AI: Build a Multi-Agent Application with CrewAI
26mAI Agents as Your Co-Developer
16mWhat Comes After AI Coding Assistants
25mBeyond the IDE: Toward Multi-Agent Orchestration
59mFrom Pair to Peer Programmer
19mBeyond agents: AI as a creative partner