The most important design decision in Wilson isn't what it can do. It's what it refuses to do on its own.
Wilson researches your supporters, maps warm paths to new donors, and drafts the outreach in your voice. It does all of that without being asked. But it will not send a single word to a single person until you've read the draft and approved it. Not the first email, not the hundredth. There is no setting to turn this off.
People sometimes assume that's a limitation we haven't gotten around to removing. It's the opposite. It's the foundation everything else is built on.
The math of a donor relationship
Here's the asymmetry that drives the whole design.
A good draft saves you fifteen minutes. A bad message costs you a relationship that took ten years to build.
That math never balances. No amount of saved time on the left side of that equation justifies risk on the right. A donor who gets a note that's wrong in a way only a machine would be wrong about, a name misread, a deceased spouse referenced, a tone that doesn't sound like you, doesn't think "their software made a mistake." They think "this organization doesn't know me." And they're right to think it, because the moment that message went out unreviewed, the organization didn't.
So the gate isn't a safety feature bolted on at the end. It's the reason the rest of the product gets to exist.
What the agent actually does all day
If nothing sends without you, what's the point of an agent at all?
The grind. The part of relationship work that eats hours and requires no judgment: reading the news for mentions of your supporters, finding the people connected to the ones who already believe in you, pulling together the background before a meeting, writing the first draft of a note you'd have written anyway three weeks later.
The agent does the work around the work. You do the relationship. Every morning it hands you a short stack: here's what I found, here's what I drafted, here's why. You approve, edit, or skip each one. The judgment, the timing, the "actually, she'd hate this," stays with the person who has it.
That's about five minutes a day. Not because the agent did five minutes of work, but because it did hours of work and then stopped at exactly the line where your knowledge of your people matters more than its speed.
Refusing is a feature
There's a version of this product that sends automatically. It would demo better. The metrics would look incredible: messages sent, touches per donor, activity per week.
It would also be a machine for spending trust you can't get back, faster than you could ever notice it happening.
I don't build that version, for Wilson or for anyone. When I build a custom agent for an organization, the approval gate comes with it, the same as it works in Wilson. If what you want is automation that runs without a human, I'm not the right fit, and I'd rather tell you that on this page than after a kickoff.
The tools are getting more capable every month. That makes the line more important, not less. The organizations that will get the most out of agents aren't the ones that hand over the keys. They're the ones that use agents to show up more prepared, more consistently, more personally, with a human deciding what actually goes out the door.
Nothing sends without you. It's the first rule I wrote, and it's not changing.