What an 85% reduction in context overhead actually means in a working sprint

The 85.5% figure in the SR-SI work is mostly about wasted orientation, not raw speed.

The more useful question is how much of a working session gets consumed by re-establishing context that should already exist.

That distinction matters because “context overhead” sounds abstract until you watch it happen inside a sprint: a developer opens a session and explains the product again, a product manager restates the state model, and someone pastes the constraints, naming conventions, ruled-out approaches, and latest scope decision.

None of that is the work; it is the preparation required before the work can begin.

In a team without context architecture, that preparation happens repeatedly. Five minutes here, ten minutes there, then the same context again when a different person starts a different task. Across a two-week sprint, the waste includes attention as much as time. People spend part of their thinking capacity recreating the conditions for useful AI output instead of using the output to move the product forward.

The SR-SI claim is that a shallow index architecture can remove most of that repeated context-maintenance burden. The persistent layer carries product identity, standing constraints, live decisions, terminology, and active priorities, while the person starting the session only adds the task-specific frame.

The useful split is simple.

Persistent context should not be retyped every session. It should live in the system, stay maintained, and remain available before the work begins.

Task context should still be written by the person doing the work. The specific feature, the specific question, the edge case being explored, and the scope of the output all need current human judgment.

The remaining 14.5% is the useful part of context variation because the work itself keeps changing.

Bigger context windows do not solve the whole problem either. A larger window can hold more, but it does not decide what deserves to be carried forward, what has expired, or which piece of context matters for the task in front of the team.

The sprint benefit is practical. Less re-briefing means faster starts, more consistent output across team members, and fewer mistakes caused by missing organizational context. The model is still reviewed, corrected, and constrained by humans, but it begins from a better orientation layer.

The SR-SI methodology post describes the architecture behind this. The AI Memory Prosthesis resource is the practical entry point for that work, not a replacement for the methodology itself.

AI does not suddenly remember like a person. The practical win is smaller and more valuable: the team stops rebuilding memory by hand every time it wants to use the tool.