Post-Velocity Software Engineering: When Change Throughput Outruns Assurance Capacity

Large language models and automation can increase software delivery throughput by lowering the effort required to propose, implement, and integrate changes. However, the capacity to assure those changes, via evidence generation, substantive review, shared understanding, and accountable approval, scales more slowly and can saturate under limited expert attention, validation budgets, and governance obligations. We hypothesize a post-velocity regime in which the rate of change approaches or exceeds assurance capacity, so justified confidence per change declines and risk shifts down- stream into operations, yielding diminishing or negative returns from further acceleration. To make this hypothesis empirically testable, we define a construct-level model relating rate of change, assurance capacity, justified confidence, and context-dependent safe velocity, and we propose trace-derived diagnostics for (i) validation lag, (ii) procedural oversight, and (iii) downstream defect shifting. We state falsifiable propositions and outline study designs (e.g., interrupted time series, difference-in-differences) that address confounds such as product maturity and reporting policy changes. We conclude by outlining an intervention space for accountability, preserving acceleration, including risk-stratified gating and differential assurance budgeting in AI-assisted workflows