Coding agents make implementation cheaper. That sounds like the whole story until you watch what happens next. The scarce thing stops being the hands that type the code and becomes the shared intent around the code. What are we building? Why this shape? Which tradeoffs have already been accepted? Who saw the browser trace, the dead end, the weird constraint, the reason the obvious version was rejected?
When implementation is slow, teams get accidental alignment as a side effect. People talk while waiting. Plans harden because the work takes time. The friction is annoying, but it gives intent somewhere to live. Once agents compress the distance between prompt and pull request, that accidental alignment disappears. The team can now manufacture divergence faster than it can notice it.
This is why the interesting frontier is not simply better coding agents. It is better shared work surfaces for agentic work. The session matters. The plan matters. The summary matters. The browser state, terminal history, failed attempt, open question, and quiet decision all matter. If they remain trapped in one person’s machine or one agent’s transcript, they are not really part of the team’s memory.
The pull request is too late to be the first shared object. By then the work already has shape, ego, dependencies, and sunk cost. Review becomes archaeology. People are not only asking whether the code is good; they are reconstructing why it exists. That is a bad use of human attention, and it gets worse as agents make the pile of completed work larger.
The durable layer has to move earlier. Teams need a place where humans and agents can gather context, shape intent, execute, revise, and leave behind enough memory that the next person or agent can re-enter the work without ritual reconstruction. Multiplayer matters, but the screen is not the main thing. Persistence is the main thing.
The deeper product is not an agent that writes code. It is an environment that preserves intention across time, people, and machines. Cheap code raises the price of weak coordination.
Sparked by Maggie Appleton’s “Zero Alignment Software Development,” a GitHub Next talk on Ace and multiplayer agent workspaces: https://maggieappleton.com/zero-alignment