# Agents need constraints — complete freedom fails > [!quote] > "Complete freedom, no constraints, just go off and do your stuff — it will fail horribly, horribly, horribly. Agents still need… [guardrails]." — Maarten Grootendorst A practitioner restatement of the vault's existing principle: safety and effectiveness come from bounding the action space, not from a better model. Current agent applications still cluster in coding + internal DevOps/data-engineering workflows. ## Related atoms - [[3. Resources/Permanent Notes/Bounded Action Spaces Not Better Models Make Agents Safe to Run]] — this talk is direct corroboration ## Links - [[Lit · AI tools accelerate WordPress feature development]] — the author's "purposefully contained" Claude Code features are this note's bounded action space applied to WordPress core work - [[Lit · AI tools democratize WordPress block development]] — handing AI block-building to non-coders raises the stakes on exactly the guardrails this talk insists on - [[3. Resources/MOCs/AI Agent Safety MOC]] — applies-in: Joins the bounded action space cluster. ## Source - [Reader: Generative AI in the Real World — Agentic Systems Fundamentals (Maarten Grootendorst)](https://read.readwise.io/read/01ktwh74dp8x882jr6th07875x) *(vault source note pending next Readwise sync)*