Repair as Representational Work: Integration Bottlenecks in AI-Assisted Development

Abstract

Research on AI-assisted programming has concentrated on the gulf of execution -- how users write successful prompts. We report a candidate phenomenon, an integration bottleneck, that lies in Norman's gulf of evaluation: a repair-relevant contribution reaches the user and fails to become actionable at the point of receipt. Two cases in an eighteen-case corpus of publicly shared AI-assisted-development accounts report this, from a peer and from the system's own output; both fall at evaluation's interpretation stage, and a third, which would fall at comparison, is reached only on an inferential reading and reported as a boundary case. A within-case contrast is consistent with actionability turning on whether the contribution can be restated as an instruction without an intervening judgement. We report this as a candidate warranting dedicated study, not an established regularity; its evidence base is retrospective author self-reports. On the submission side, medium alone did not sort the corpus contrasts; immediate uptake aligned with a checkable condition across four decisive contrasts, while a fifth case shows checkability sufficient for immediate uptake did not guarantee persistence. A twenty-trial multi-turn probe across two current models observed no constraint loss in fifteen judgeable narrow trials, and a ten-trial extension with an explicit restructuring request none in eight. Study 2 is not a replication attempt: it isolates regeneration, the mechanism the corpus authors invoke, and removes it as a sufficient explanation. We specify four interaction requirements for an accepted-constraint ledger; a conformance analysis of twelve mechanisms found none documented to satisfy all four.