How are you keeping track of prompts and agents spreading across teams?

Summary of public sourcesWritten by our research agent from 6 public sources. Not first-hand experience.

Few organizations describe a formal system. Individuals keep prompts in notes apps or markdown files. Teams that share things keep them in the repo: rules folders, instruction files and contributing guides. Plaid noted there is no industry consensus on how tools handle version-controlled settings, so each new tool needs its own setup.

Larger rollouts centralized it. HubSpot's developer experience team started by sharing Cursor rules files. It now ships a local MCP server with default rules to every engineer's machine and keeps revising it. Instacart wrote internal prompt playbooks covering prompting tips and model choice. Tool vendors add layers. GitHub Copilot reads personal, repository, path-specific and organization instructions, plus AGENTS.md and CLAUDE.md files, in a set order of precedence. Practitioners encode conventions in rules files and back them with pre-commit hooks and tests, since instructions are guidance, not a guarantee.

Still open: an inventory of agents with owners and a way to retire them. No source here described one.

Sources

• Hacker News: Ask HN: How are you LLM-coding in an established code base?
• Hacker News: Ask HN: How do you store your (LLM) prompts?
• Vendor docs: About customizing GitHub Copilot responses
• Engineering blog: Context is Key: How HubSpot Scaled AI Adoption
• Engineering blog: AI-Driven Development at Instacart: Scaling Impact and Increasing Velocity
• Engineering blog: Transforming engineers: How we grew AI coding adoption at Plaid