What an AI outbound agent should never do
Eleven rules FOUND's agent keeps whatever the settings say, the mistake each one prevents, and why they live in the code that sends rather than in a prompt.
On this page 8 sections
The question to ask of software that writes in your name isn't what it can do. It's what it refuses to do on a bad day: when the list is thin, when a reply is angry, when the week has been long.
These are FOUND's refusals. Each one exists because the opposite happens often enough in outbound to cost someone an account, a domain or a deal, and each one is enforced where the sending happens, for every agent on every account.
Rules, not settings
A setting is something you can forget to set. An instruction in a prompt is something a model can ignore, and eventually will. So the rules below are neither. They sit in the code every invitation, message and email passes through, the same pacer and the same caps for every agent, and a message that breaks one is held back rather than sent.
You can make FOUND slower or more careful than this. You can't make it less.
Pacing: never faster than a careful person
Never send in a new account's first 48 hours
LinkedIn watches a newly connected account closely. An account that connects a tool and starts inviting the same afternoon is the pattern most worth avoiding, so for 48 hours the agent only finds people. You get leads the day you connect and messages from day three.
Never go past the limits
35 invitations a week to start, rising one step every two clean weeks, and never more than 25 in a day: 125 a week on the Monday to Friday window every account starts with. Up to 100 messages and 100 profile visits a week. A mailbox warms from 5 emails a day on its first day of sending, 5 more each day to 30, then 40 and 60. The full ladder is in our pacing guide.
Never send on a tick, and never catch up
LinkedIn actions go out inside your window, on the days you choose, at least 90 seconds apart plus a random wait of up to four minutes. A day the window missed is gone: tomorrow doesn't send double. If LinkedIn pushes back, the account stops for 72 hours and drops a step.
Who it writes to
Never write again to someone who said stop
A reply of stop, unsubscribe or not interested stops that person at once, and anything already queued to them is cancelled. Nobody has to notice the reply first.
Never reach anyone on your block list
Block a company or a domain and nobody there gets an invitation, a message or an email from any of your agents; anything already planned for them is cancelled. A domain covers its subdomains and every address at it, so your customers, your investors and your old employer stay out of it.
Never keep going with someone you marked a poor fit
Press Fewer like this on a lead and the agent stops contacting them straight away, cancels what was queued, and uses the example to judge the next people it finds.
What it writes
Never say "I saw you liked my post"
The writer is told never to say the reader liked, commented on, reacted to or engaged with anything. Being watched is the creepiest thing a first message can admit to. It builds on the interest instead, in plain words, and the first message pitches no features, names no price and carries no link.
Never send a message with a blank in it
If a message still holds a field it couldn't fill, a leftover in square brackets or braces, it doesn't go. "Hi [FirstName]" reaching a real person is the fastest way to show them a machine wrote it.
Replies
Never answer on its own unless you asked it to
A reply stops the sequence. With drafted replies switched on, it also gets a drafted answer, which goes only when you press Send. The agent answers by itself only where you switched it on, one conversation or every new one on an account, which is off by default. It hands you anything only you can answer, and once you reply on a thread yourself, the thread is yours: the agent never writes on it again.
Never touch your calendar, or pick a time
FOUND offers your meeting link to people who are interested. It never reads or writes your calendar, and never proposes or accepts a specific time: the prospect picks a free slot on your own booking tool. More on that in turning replies into booked calls.
Your accounts and your money
Never keep your password, or spend without saying so
You connect LinkedIn on FOUND's own page, and your password passes through our server once, straight to our connection provider. It is never stored or logged. Credits work the same way: a work email costs a credit only when one is found, and Ask shows the most it can cost before it runs and gives back what it didn't use.
What stays with you
Rules keep an agent from doing damage. They don't make its judgment yours. Who counts as your buyer, what your brief says, whether review mode stays on and what you say on the call are still your calls. Nobody can promise LinkedIn will never act on an account, as our terms say plainly; pacing is how FOUND keeps that risk low.
If you're weighing any agent, ask to see its refusals in writing. Ours are on the demo page, and the rest of what it does is in our guide to AI SDRs.
Questions
Can I switch these rules off?
No. You can make the agent slower or more careful, with review mode, a lower daily limit on your mailbox, or by pausing a signal or the whole agent, but the limits and refusals are enforced in code for every account.
Does review mode cover the agent's answers too?
Yes. With review mode on, every invitation, message, email and answer from the agent waits for your approval, and you can edit any of it first.
What happens if LinkedIn pushes back?
The account stops for 72 hours and drops one step on its ladder. It can't climb again while it has a failed send in the last 14 days.
What if someone wants no FOUND customer to contact them?
They can ask through our opt-out page. Replying stop reaches the business that wrote to them at once; a block for every FOUND customer is applied within 30 days.
Someone contacted you through FOUND? Opt out here.