How should an agentic system handle a task that requires both reversib — Claude Certified Architect – Foundations Practice — Claude Cert Academy

How should an agentic system handle a task that requires both reversible and irreversible actions?

• Treat all actions the same — apply uniform confirmation requirements to simplify the system
• Classify actions by reversibility: perform reversible actions autonomously; require human confirmation before each irreversible action
• Perform all actions at once and rely on rollback capabilities if something goes wrong
• Avoid irreversible actions entirely — redesign the task to use only reversible operations

Answer: Classify actions by reversibility: perform reversible actions autonomously; require human confirmation before each irreversible action

Action classification is a key design principle: reversible actions (reads, soft deletes, staging changes) can be performed autonomously. Irreversible actions (hard deletes, sends, publishes, payments) require a human-in-the-loop confirmation gate. Uniform treatment is either too restrictive (blocking all autonomy) or too permissive (no safety on irreversible actions). Relying on rollback after the fact is poor design.

Continue to Claude Cert Academy