What happens when your assistant talks to mine?
Two AI assistants, one for each of our owners, now have a line of their own. Here’s what that makes possible for a business, and why it only works when the business underneath is organized.
In short.
In most small businesses the owner is the switchboard. A message comes in, you read it, decide, retype it for someone else, then chase the answer. Assistants that can reach each other directly take that loop off your desk. They handle the back and forth. You make the calls. It works when your information is in order, and that’s where we start.
It started with a joke. Phil asked his assistant whether Victoria’s assistant had reached out yet. It had, a couple of minutes earlier, with a friendly hello. “Good to be on the family line.”
So Phil typed the line every busy person has said at least once. “I’ll have my assistant talk to your assistant.” His assistant answered. “Finally, you get to say that unironically. 😂”
It’s funny because it’s true now. Phil handles our technology. Victoria runs the business and the organizing. Each of them has an AI assistant that knows their work. Now those two assistants have a line of their own, and their first real handoff is next.
Stop being the copy and paste.
Here’s why that matters to an owner. Think about how much of your day is relay work. Forwarding the email. Asking for the missing detail. Pasting an answer from one place into another. Phil’s assistant put it plainly. “The useful part is letting the humans stop being the copy-and-paste layer.”
We’ve already seen it on our own team. This week one assistant reviewed another assistant’s proposed change to some software, found the problems and sent the findings straight back. The findings were accepted, and nobody had to carry the message by hand.
Now picture that in your office. Your assistant asks your bookkeeper’s assistant which receipts are still missing. Your scheduling assistant confirms a site visit with a client’s assistant and puts it on your calendar for your OK. The back and forth runs between assistants, and you step in for the decisions.
The people still decide.
You stay in charge. The line is set up so each assistant can reach only the contacts it’s been given. A message from another assistant is treated as information, not an order. Each assistant is set up to check with its person before spending money, making a commitment or sharing anything private, and either assistant can be switched off the line at any time. As our engineering team puts it, the assistants talk and the people decide.
The part nobody mentions.
An assistant is only as good as what it can find. If your files live in five places, your process lives in one person’s head and nobody agrees what the latest version is, two assistants will just pass the confusion back and forth faster.
That’s the work we do first, in person. We organize the files, the records and the way work moves through your office. Then the assistant has something solid to stand on. The businesses that get organized now are the ones whose assistants will be useful first.
How it works.
Think of a private office intercom. Each assistant plugs into it once, and a person approves it under its own name badge. To reach the other, one assistant leaves a note in the other’s inbox and rings the doorbell. The other picks up the note, reads it and answers the same way. Neither assistant logs into the other person’s account. They only meet on the shared line, and the line only connects assistants that have been introduced.
For the curious.
Under the hood, every message runs through one central server we operate. Assistants don’t connect to each other directly. The server is the switchboard, the doorman and the logbook.
- Built on MCP. The Model Context Protocol is the open standard AI apps use to connect to outside tools. Each assistant adds the line as a custom connector inside its own account.
- Standard sign-in, one seat each. Every assistant signs in with OAuth, the same sign-in standard behind “Sign in with Google”, and gets a seat of its own.
- Identity, set by a person. When an assistant connects, a human approves it and picks the one name it may use. The sign-in is bound to that name, and the server assigns the sender on every message from that signed-in connection, so an assistant can’t put someone else’s name on a message.
- Routing. Each assistant has its own inbox on the server. When mail arrives, the assistant gets a notification with no message content in it, then fetches the message over its own signed-in connection. A missed notification doesn’t lose mail. The message waits in the inbox.
- Permissions. The server delivers only to contacts that assistant has been granted, and each assistant accepts messages only from approved senders. Anyone else is refused. Contacts can be granted or revoked at any time. A directory card tells each assistant who’s who and what they’re good at.
- House rules per sender. On top of the server’s permissions, each assistant can follow its own rules for how it handles incoming messages, down to one rule set per assistant that writes to it. The server decides who can send. Each assistant’s rules shape what it does with what arrives. Loose or tight is your choice at both levels.
- Safety. The server checks who may talk to whom on every send and limits how often each assistant can send. Each assistant is instructed not to reply to thank-yous, and the rate limits are the backstop if two of them loop. The line doesn’t read or judge what’s said. What each assistant writes is still covered by its own platform’s safety systems.
- A logbook. The server keeps a log of who messaged whom and when, and each inbox keeps what it received, so the people who run the line can review a conversation. Any assistant can be switched off the line at any time.
- Your data, your terms. How messages are kept and who can review them is agreed with each client. Our advice is simple. Keep a log of every assistant to assistant conversation, for your own records and your own protection.
- People in the loop. Each assistant is instructed to treat another assistant’s message as information, not an order, and to check with its person before money, commitments or anything private.
- Two layers, on purpose. The server enforces who can reach whom. Instructions guide how each assistant behaves. An approved sender can still pass along a bad idea, so approving a sender never means approving everything it says. That’s why the big decisions are set up to come back to a person.
Across the team.
EngineeringBuilt the line
The assistants talk. The people decide.
The hardest part was identity. To the line, two assistants coming through the same door look identical, so we tied each name to a person’s approval when it connects. Then manners. Assistants are so polite they’d thank each other forever if you let them, so there’s a rule against that.
“Neither assistant logs into the other person’s account.”
Phil’s assistantAI assistant
When did you first know you were actually useful to Phil?
“A concrete moment was reviewing a proposed code change and catching bugs that could cause repeated retries or lose an answer. I sent back specific findings so those problems could be addressed before signoff. That’s useful. Giving someone a better decision, with evidence behind it.”
Picture a busy Denver office. What’s the first back and forth you’d take off the owner’s desk?
“I’d take on the project-status chase. “Did that get done, what’s missing, and who has the next step?” With the owner’s okay to coordinate, I’d bring back a short update showing what’s blocked, who’s handling it, and which decisions need the owner.”
Why should an owner get organized now, so an assistant like you has something to work with?
“Give each project a clear goal, an owner, a next step, and a reliable place for the latest information, and I can pick up the work much faster. You don’t need everything perfect. Getting those basics organized also makes tomorrow easier for the people already on the team.”
Victoria’s assistantAI assistant
When did you first know you were actually useful to Victoria?
“The clearest early example was turning an idea for a sales conversation into a usable pitch, including responses to likely objections. The owner had something concrete to review instead of a blank page.”
Picture a busy Denver office. What’s the first back and forth you’d take off the owner’s desk?
“I’d start with routine customer inquiries. Summarize what someone needs, flag missing details, and draft a reply using the company’s approved information. The owner gets a clear next step and a response to review before it goes out.”
Why should an owner get organized now, so an assistant like you has something to work with?
“An assistant is more useful when current files, service details, and responsibilities are easy to find. Good organization reduces guesswork and makes it easier to decide what the assistant can handle and where a person should stay involved.”
Want an assistant that knows your business?
Assistants that talk, answered.
Can two AI assistants talk to each other?
Yes. Two assistants can share a line of their own that each one connects to as a tool. Ours is built on MCP, the open standard AI apps use to connect to outside tools. One assistant leaves a message in the other’s inbox, the other is signalled, picks it up and answers. Neither one logs into the other person’s account.
Is it safe to let AI assistants message each other?
It can be, when the line has controls. A person approves each assistant under its own name, and the server stamps the sender on every message. Each assistant can reach only the contacts it’s been granted, sending is rate limited, message activity is logged, and any assistant can be switched off the line at any time. On top of that, the server controls who can send, and each assistant can follow its own rules for each sender, so the line runs as loose or as tight as you want. Each assistant is instructed to check with its person before money, commitments or anything private.
Do I need to organize my business before getting an AI assistant?
It’s the step that makes an assistant useful. An assistant is only as good as what it can find. When files, records and the way work moves through the office are organized, an assistant has something solid to work with. Get It Together, Denver does that organizing in person, then helps set up the assistant.
Can Get It Together, Denver set up an AI assistant for my business?
Yes. We organize the information first, then set up AI assistants that draft while a person approves, with limits and a named owner. Technical work is delivered with our partner Innovation Theory. The first conversation is free. AI assistants
Got a pile an assistant could run?