NNGroup: The custodial era of UX — cleaning up after AI
Anna Kaley and Raluca Budiu at Nielsen Norman Group published this piece on August 28, 2026, coining a term that resonates with almost every design team that has adopted AI tooling: the custodial era. The idea is simple: AI now lets product teams generate content, prototype, and build working features far faster than UX can evaluate them, and the result is a backlog of unreviewed AI output that reaches users before it has been properly assessed. The authors call this accumulation “UX debt,” and they argue that addressing it is the most pressing design challenge of the current moment.
The article describes a cycle that will feel familiar. A new idea arrives, AI production turns it into a working prototype or live feature within hours, the evaluation step gets compressed or skipped, and users encounter confusing interfaces, poor adoption rates, or outright confusion. Designers then spend significant time cleaning up problems that a structured review process would have caught before release. This is the custodial pattern — and the authors’ point is not that it is shameful, but that it is now the primary shape of UX work at many companies.
Kaley and Budiu propose three strategic responses. The first is building shared judgment across the team about what actually deserves to be built. When everyone can generate UI and code, the question of whether something should exist at all becomes harder to surface. Design needs to hold that standard even when speed is the dominant pressure. The second response is accelerating UX evaluation itself — faster research methods, lighter evaluation templates, instruments that can run at the pace AI generates outputs. The third is integrating UX guidance into the AI generation tools directly, so that predictable problems get caught at the source rather than during cleanup.
The article also reframes the custodial role more generously. Cleanup work carries information: it shows which patterns AI produces that users consistently find confusing, which design decisions AI cannot make without context, and where human judgment is irreplaceable. Treating that information as data for the team — rather than as embarrassing evidence of a flawed process — turns custodial work into a way to improve both the AI tooling and the team’s shared standards over time.
Useful for design leads and senior individual contributors at teams where AI-assisted development has moved faster than evaluation capacity. The framing is honest about the difficulty without recommending a retreat from AI tooling, which makes it more actionable than most takes on the subject.