AI agent guardrails compared
Guardrails for AI agents come in three kinds. Here's what each one controls, who it's for, and where they fit together.
By Jude Eric ·
1. Policy engines and MCP gateways (for developers)
Open-source projects such as PolicyLayer, AgentGuard, controlzero and Agent Policy sit between an agent and its tools and check each tool call against rules written in YAML: allow, deny, or ask a person. Useful ideas from them: rate limits that catch runaway loops, stricter rules when an agent runs unattended, and plain explanations returned with a denial. They're installed and written by developers, for one agent or one team.
2. Card and wallet limits (from payment companies)
Visa, Mastercard, Stripe Link, Privacy.com and agent-wallet providers let you give an agent a card or wallet with limits: an amount, approved merchants, a time window, or approval per payment. The bank enforces it, so an agent can't skip it. It only knows amounts and merchants, though, not what is being bought or why.
3. Rules for people (YeNo)
YeNo is one set of rules every AI a person uses checks before acting: a monthly budget across all of them, a never list, trusted shops, limits per AI, rules for a child's AI, and rules learned from real AI mistakes (asked to find isn't asked to buy; no more than 3 of one thing; nothing during quiet hours). It answers yes, no with the reason, or asks the owner's phone. Approvals live on the phone, out of the AI's reach, and an AI that tries to raise its own limit gets a no.
How they fit together
| Policy gateways | Card limits | YeNo | |
|---|---|---|---|
| Who sets it up | Developers | Anyone with the card | Anyone, on their phone |
| Knows what's being bought | If written into rules | No (amount, merchant) | Yes |
| Across every AI you use | Per deployment | If they all use that card | Yes |
| Can the agent skip it? | Not if all tools go through it | No | Not for actions through YeNo; elsewhere it's asked to check |
The strongest setup uses two: a card with its own limit (hard stop) plus YeNo's rules (what, where, why, across every AI). Builders can also make promises YeNo enforces and show the "Works with YeNo" badge: see YeNo for bot builders.
Rules your users' AIs already follow
Call check_rules before acting (MCP or plain HTTP), and declare promises YeNo enforces.
Read the bot builder docsCite this page
Jude Eric, YeNo Rules. “AI agent guardrails compared.” Last updated . www.yenorules.com/docs/ai-agent-guardrails