Agent builder

When someone needs an AI system to act autonomously on their behalf, let them define exactly which tools it can use and what rules bound it. Set the limits before it acts, not after.

Relevant principles

Safety is imperative
High autonomy & human control
Control not input

What the research says

Justified by the global controls principle: letting users define what the AI does, and when, is control, and control is one of the more reliable antecedents of trust. There is no pattern-specific trust experiment here, so treat it as principle-backed rather than measured. 1 2

Interactive demo

Email assistant

Off

When an email arrives from , using , and .

Preview

Priya Shah08:12Can we move Thursday’s shoot?Something’s come up. Any chance of Friday instead?

Draft

“Friday at 10 works. See you then.”