The three things the agent is not allowed to do without us
Spending money, contacting a real person, touching anything live. All three were written down while the software could not yet do any of them.
August 2026 · 2 min read
The prospecting agent runs on its own. It sources businesses, researches them, scores them and rejects most of them without anyone watching.
It stops dead before three things: spending money, contacting a real person, and touching anything live.
All three were written down before the agent could do any of them.
Why the timing is the whole point
Once a capability exists, every argument for restraining it starts to sound like an argument against shipping. The restraint gets deferred, then written after the first incident, by which point it is a remedy rather than a design decision and somebody has already been affected.
Deciding while the capability is still hypothetical costs nothing and is the only moment the decision is genuinely free.
On the second one particularly
No message goes to a real human being until a person has read the exact text and said send. Not a template approved once. The actual words, each time.