- Author
- Jenny Y. Huang
- Published
- 4 May 2026
- Kind
- Personal research blog essay, with two prototype interfaces
- Original
- jennyhuang19.github.io/slow-ai-ai-that-meets-a-humans-pace/
- Subject
- Coding with agents
What it is
Jenny Y. Huang writes on a personal site that lists publications and news. The essay draws on the author's own prior work and conference talks. It opens with a story about learning math.
What it says
- Huang says fluent, fast AI output makes accepting answers at face value the easy path.
- Huang proposes interfaces that show the decisions behind a response, as an interactive decision tree.
- Huang says study participants felt more ownership with the decision tree than with linear chat.
- Huang suggests presenting answers as deliberations among agents, so readers act as critics.
- Huang says long chats mislead models. One figure shows GPT-5.2 reusing a UMAP metric in t-SNE code.
- Huang warns that fast AI culture reinforces itself. People should choose when to slow down.
What evidence it gives
The essay cites the author's own study and two prototypes. One is a decision tree and the other a browser extension for setting AI boundaries. It shows one figure of a model error. It gives no sample size or statistics.
Interests and limits
The author's own group built the prototypes and ran the study. The study details are not given on the page. The essay cites other work on AI risks without reviewing it. The size of the effect is not stated.