# Keep responsibility human

Source: https://segern.com/practice/keep-responsibility-human

Delegate the work. Keep the responsibility.

Delegation is useful when it expands what can be done without making the outcome harder to understand or trust. That applies to people, software, and AI.

Before handing over work, we make the purpose, boundaries, and expected evidence clear. For AI, this includes what it may access or change, what it must not do, when it should stop, and which actions need human approval. Confidence in an answer is not permission to act on it.

We distinguish a plausible suggestion from a verified result, and both from an authorized decision. Consequential work has a named human who accepts responsibility for the decision and the obligations it creates. Contributors should not have to guess who can resolve a disagreement or approve a change.

Using several agents does not by itself create independent verification. Their work still needs checks appropriate to the consequence, and someone accountable for responding when those checks fail.

In an advisory relationship, our role is equally explicit. We are responsible for the quality and clarity of our advice and the work we agree to carry out. The founder retains the decisions that belong to their company unless a specific responsibility has been expressly delegated.

*Who can decide, what may be delegated, and what evidence is required before acting?*