The Perfection Paradox: From Architect to Curator in AI-Assisted API Design

Mak Ahmad; Andrew Macvean; JJ Geewax; David Karger

cs.SEMar 12, 2026

The Perfection Paradox: From Architect to Curator in AI-Assisted API Design

Authors: Mak Ahmad, Andrew Macvean, JJ Geewax, David Karger

Organizations: Google, San Francisco, CA, USA · Google, Seattle, USA · Google, Singapore · MIT, Cambridge, MA, USA

Abstract

Enterprise API design is often bottlenecked by the tension between rapid feature delivery and the rigorous maintenance of usability standards. We present an industrial case study evaluating an AI-assisted design workflow trained on API Improvement Proposals (AIPs). Through a controlled study with 16 industry experts, we compared AI-generated API specifications against human-authored ones. While quantitative results indicated AI superiority in 10 of 11 usability dimensions and an 87% reduction in authoring time, qualitative analysis revealed a paradox: experts frequently misidentified AI work as human (19% accuracy) yet described the designs as unsettlingly "perfect." We characterize this as a "Perfection Paradox" -- where hyper-consistency signals a lack of pragmatic human judgment. We discuss the implications of this perfection paradox, proposing a shift in the human designer's role from the "drafter" of specifications to the "curator" of AI-generated patterns.

Figures & tables

Figure 1. AI-Enhanced API Design Workflow: From Requirements to Final Specifications This flowchart illustrates the AI-enhanced API design process. It consists of five main stages: Requirements Input, AI-Assisted Designer, Generated API Specification, Expert Evaluation, and Iterative Refinement. The process flows from top to bottom, with arrows indicating the sequence. A feedback loop connects Iterative Refinement back to the AI-Assisted Designer.

Characteristic	Distribution
Programming Experience	3–5y: 1, 6–10y: 2, 11–15y: 4, 16–20y: 2, 21+y: 7
API Design Experience	1–2y: 1, 3–5y: 3, 6–10y: 4, 11–15y: 8
AIP Familiarity	Expert: 7, Very familiar: 2, Moderate: 1, Slight: 3, None: 3
AI Tool Experience	Extensive: 5, Moderate: 7, Limited: 4

Table 1. Participant Demographics (N=16)

Dimension	Human (A)	AI (B)	Human (C)
Abstraction Level	3.00	3.38	3.13
Work-Step Unit	3.06	3.19	3.19
Working Framework	3.19	3.50	3.13
Progressive Evaluation	3.25	3.31	3.00
Premature Commitment	2.88	3.25	2.63
Penetrability	3.06	3.75	2.69

Table 2. Mean Scores Across Cognitive Dimensions

Software Engineering

Software engineering studies how to design, build, test, deploy, and maintain reliable software systems and the processes behind them.

See all 270 papers

When AI Designs AI: Innovation or Imitation?
Aug 18, 2026Yikang Yang, Zhengxin Yang, Luzhou Peng +4LLM Agents Automated Algorithm Discovery
State Key Lab of Processors, Institute of Computing Technology, CAS, Beijing 100190, China · Institute of Computing Technology, Chinese Academy of Sciences · University of Chinese Academy of Sciences +2
Vibe to Code: Elucidating Strategic Oscillation of Tacit Knowledge in Generative AI Design Workflows -- An Exploratory Qualitative Study
Jul 25, 2026Daisaku SatoHuman-AI Interaction Human-AI Co-Creation
Graduate School of Knowledge Science, Japan Advanced Institute of Science and Technology · (JAIST) / Sumitomo Mitsui Finance and Leasing Co., Ltd.
Design Generative AI for Practitioners: Exploring Interaction Approaches Aligned with Creative Practice
Mar 3, 2026Xiaohan Peng, Wendy E. Mackay, Janin KochMixed-Initiative Interaction Human-AI Interaction
LISN, Université Paris-Saclay, CNRS, Inria, Orsay, France · UMR 9189 CRIStAL, Univ. Lille, Inria, CNRS, Centrale Lille, Lille, France

Aug 18, 2026J/K move · Enter open · S save

When AI Designs AI: Innovation or Imitation?

Yikang Yang, Zhengxin Yang, Luzhou Peng, Minghao Luo+3

State Key Lab of Processors, Institute of Computing Technology, CAS, Beijing 100190, China · Institute of Computing Technology, Chinese Academy of Sciences · University of Chinese Academy of Sciences+2

LLM Agents Automated Algorithm Discovery LLM Agent Evaluation

Open paper AskPDF arXiv

Recent advances in LLM agents have made them increasingly capable of designing methods for complex AI tasks. This raises two central questions about agent-designed methods relative to human-designed methods: how well they perform, and how different their algorithmic designs are. To study these questions, this paper introduces an analysis that derives task-specific algorithmic design spaces from human-designed methods, maps both human- and agent-designed methods into these spaces, and quantifies their algorithmic differences at the module level. Widely used LLM agents are evaluated on a suite of representative, open-ended AI tasks spanning multiple modalities, and the methods they design are analyzed in terms of both task performance and algorithmic differences from human-designed methods. Experimental results show that current agents can occasionally match or surpass human state-of-the-art (SOTA) performance (10/72 configurations), but such success does not generalize reliably across tasks or agents. Moreover, 96.8% of agent-designed methods fall within human-derived algorithmic design spaces, largely recombining algorithmic choices found in human-designed methods, while nearly half exactly match an existing human algorithmic design. Taken together, these findings suggest that although current agents can occasionally match or surpass human SOTA performance, their algorithmic designs remain within human-derived algorithmic design spaces, reflecting the reuse and recombination of algorithmic choices.

Abstract

Software Engineering

Explore similar work

When AI Designs AI: Innovation or Imitation?

Vibe to Code: Elucidating Strategic Oscillation of Tacit Knowledge in Generative AI Design Workflows -- An Exploratory Qualitative Study

Design Generative AI for Practitioners: Exploring Interaction Approaches Aligned with Creative Practice