---
title: "You can't whisper at an AI agent"
description: "Stripe engineers ran roughly a dozen experiments to determine how to get AI agents to use current API versions, follow integration best practices, and load the Stripe skill. Passive hints, including SDK files, AGENTS.md guidance, and API warning responses, were largely ignored, while active prompts and distribution mechanisms performed better. A modular, progressively disclosed skill structure outperformed a monolithic file by roughly 10% and reduced token usage; a CLI login prompt led 30–35% of viewers to copy its install command, and explicit compatibility-mode errors prompted agents to correct version mismatches. The results suggest that agent-facing developer experience is primarily a distribution problem, with hard steering through errors and loaded instructions more reliable than soft hints agents may never encounter."
---

# You can't whisper at an AI agent

[Stripe](https://yomu.fyi/company/stripe) · James Beswick · May 14, 2026

**Type:** Explainer

## Summary

Stripe engineers ran roughly a dozen experiments to determine how to get AI agents to use current API versions, follow integration best practices, and load the Stripe skill. Passive hints, including SDK files, AGENTS.md guidance, and API warning responses, were largely ignored, while active prompts and distribution mechanisms performed better. A modular, progressively disclosed skill structure outperformed a monolithic file by roughly 10% and reduced token usage; a CLI login prompt led 30–35% of viewers to copy its install command, and explicit compatibility-mode errors prompted agents to correct version mismatches. The results suggest that agent-facing developer experience is primarily a distribution problem, with hard steering through errors and loaded instructions more reliable than soft hints agents may never encounter.

## Context

Stripe investigated why agents integrating with older APIs were making decisions contrary to recommended best practices without users knowing. The work focused on whether agents would discover guidance through code, API responses, documentation, skills, and onboarding flows.

## Approach / What changed

The team ran roughly a dozen experiments across passive hints, active prompts, and distribution plays. It modified SDK packages and API responses, restructured skill files for progressive disclosure, added CLI and authentication prompts, placed skills on documentation pages, and tested explicit errors for outdated API versions.

## Takeaways

- Agents almost never read files inside dependency directories and ignored warning data in API responses, including compatibility-mode skill-install instructions.
- A modular skill structure outperformed a monolithic file by roughly 10% across the evaluation suite while reducing token usage and allowing product teams to maintain separate sub-skills.
- Explicit errors reliably triggered correction of API version mismatches in testing, unlike warnings; 30–35% of users shown the CLI login prompt copied its install command.

**Tags:** [Developer Experience](https://yomu.fyi/topic/developer-experience), [LLMs](https://yomu.fyi/topic/llm), [REST APIs](https://yomu.fyi/topic/rest-api), [Testing](https://yomu.fyi/topic/testing)

- Source: [Stripe](https://stripe.dev/blog/ai-steering-experiments)
- Source URL: https://stripe.dev/blog/ai-steering-experiments
- Ingested by Yomu: 2026-08-29T00:00:55.225Z

[Read original post](https://stripe.dev/blog/ai-steering-experiments)