🧠

Guardrails, Permissions & Safety Architecture

The moment an agent can spend money, engineering discipline is the product.

AI-Native Operator's Field Manual · Expert · 8 min read · 5 quiz questions

By Ronan Liedmeier, fractional head of growth and course author · Reviewed 13 October 2026

What you will learn

  • The permission matrix
  • The non-negotiables checklist
  • Sell it, don't hide it

Lesson preview

Every tool an agent can touch falls into one of five risk classes: readonly, reversible-write, financial, destructive, or external-comms. Each class gets its own policy. The golden rule: enforcement lives in hooks and the gateway — deterministic code — never only in the system prompt. A prompt is a preference. A hook is a law. Think of it this way: the model is the pilot. The harness is the flight envelope. The pilot can't override the envelope. Here's how to handle each risk class. readonly — free use, but always logged. Looking is fine; leaving a trace is required. reversible-write (adding labels, creating drafts, adding negatives) — dry-run first, show the diff, get human approval, then execute and log. The diff is the approval UI. financial (budget changes, bid adjustments) — requires explicit approval plus hard caps enforced in code: both a percentage limit and an absolute monetary ceiling. These limits live in the gateway, not in the prompt. destructive (delete campaign, delete audience) — not exposed to agents at all. Human-only, ever. There is no careful way to give an agent a delete button. external-comms (send email, publish post) — approve per item at lower rungs; per batch with sampled review at higher autonomy levels.

Continue reading

Key terms in this module

AI guardrails
AI guardrails are checks and limits placed around an AI system that block or flag unsafe or wrong inputs and outputs before they cause harm. The strongest ones run as code outside the model, so a cleverly worded request can't talk its way past them.
Human in the loop
Human in the loop is a setup where an AI system pauses at set points so a person can approve or correct its work before it takes effect. It is used for actions that are hard to undo, such as sending emails to customers or moving money.
Prompt injection
Prompt injection is an attack where text an AI reads, such as an email, web page or file, contains instructions that make the model do something its user did not intend. OWASP ranks it first in its Top 10 risks for LLM applications.

Frequently asked questions

Who is "Guardrails, Permissions & Safety Architecture" for?
This Expert module in AI-Native Operator's Field Manual is for operators, founders and growth teams who run AI in real work and want production-grade judgement. It takes about 8 minutes and ends with a 5-question quiz.
Do I need to code?
Not necessarily. The workflows use operator tools such as Claude, n8n and Lovable; some exercises are easier if you are comfortable editing a config file or prompt template.
What comes before and after this module?
Before "Guardrails, Permissions & Safety Architecture", take "Evals & Review Loops" (module 17 of AI-Native Operator's Field Manual). Afterwards, continue with "Building Products Fast".

Unlock this lesson and the full academy

  • Monthly€9.99 / month
  • Yearly€69.99 / year
  • Lifetime€129.99 one time
See plansStart with the free lessons

Loading your lesson…

Related modules