Coding agents solved the wrong problem.
Not entirely. Cursor, Claude Code, and Codex made individual engineers dramatically faster at writing code. That's real. But here's what nobody talks about: the bottleneck in software development was never writing code.
The bottleneck is coordination.
The coordination problem
A PM has an idea for a feature. Here's what happens today:
- PM writes a ticket
- Ticket goes into the backlog
- Ticket gets prioritized in sprint planning
- Engineer picks it up 2 weeks later
- Engineer asks PM what they meant
- PM clarifies
- Engineer builds it
- QA tests it
- PM reviews and asks for changes
- Another cycle of back and forth
- Feature ships, 3-4 weeks after the original idea
This is the coordination problem. AI made the individual faster. It didn't make the team faster.
What multiplayer development looks like
The future isn't faster agents. It's the whole team participating in the development process:
PMs describe what they want, not what to build. Instead of writing a spec that an engineer interprets, a PM describes the outcome they want. The system reads the codebase, surfaces the trade-offs, and shows visual options. The PM picks one. The agent builds it.
Engineers review, they don't translate. Today, engineers spend half their time translating PM requests into code. In a multiplayer model, the translation happens automatically. Engineers focus on architecture, code quality, and system design. They review and approve, not translate and implement.
Everyone sees the same thing. When a feature is being built, the whole team can see the plan, the trade-offs considered, the agent's progress, and a live preview of the result. No more "I thought you meant..." conversations.
Preview, don't describe. Instead of describing what changed in a PR description, every change gets a live URL. Click through it. See the feature working. Give feedback on the real thing, not a screenshot or a description.
Why this hasn't happened before
Three things had to be true for multiplayer development to work:
- AI had to be good enough to write production code. It is now.
- Preview environments had to be instant and cheap. They're getting there.
- Someone had to build the coordination layer. This is the hard part.
Without the coordination layer, you have a team of individuals using AI separately. Each person is faster, but the team isn't.
The multiplier effect
When the whole team ships together, the math changes:
- A PM idea goes from concept to live preview in hours, not weeks
- An engineer reviews 10 agent-built PRs instead of writing 3 features
- A designer sees their changes live, in the real product, before anyone merges
- A founder ships a landing page update without asking engineering for help
The teams that figure this out first will ship 10x more product than their competitors. Not because their engineers are faster. Because their whole team is shipping.