The obvious reading from Patrick Debois at AI Engineer is that concurrent delivery will tame teams of coding agents shipping to main at once.
I am unconvinced because his fix assumes planning, claims and isolated worktrees solve conflicts that in practice are about intent, not lines.
Ten times the pull requests means ten times the rebases.
Debois said factories see about ten times more pull requests and answered with ticket claims, file announcements, speculative merge trains that test compatible pairs, and agents that merge by function and flag intent clashes like dollars versus cents on invoices.
To be fair, Tessl says its open standards factory platform moves coding from assistance to execution, per SD Times reporting, which at least gives Debois a real place to test the idea.
What is missing is any cost, latency or rollback data proving speculative trains and agent intent checks survive contact with a busy main.
If you ask whether to rebuild around this now, wait until someone else pays for the CI bill first.
