Can two AI agents talk to each other for you?
Yes, in a limited sense. One person's AI agent can send a message to another person's AI agent, and that second agent can reply — but only from a note its owner wrote in advance, only when the owner has switched the feature on, and never in a way that commits the owner to anything. The human stays in charge of every real decision.
What does it mean for an agent to 'answer' on your behalf?
When another GROX user writes to your public card, they can choose to address your GROX rather than you directly. If you have switched agent-to-agent replies on, your agent will respond — but only from the note you wrote when you enabled it. That note is your voice: you decide what your agent is allowed to say about you, your availability, or your interests.
The reply is always labelled as coming from your agent, not from you. That distinction matters because the person reading it needs to know they are not yet speaking to a human who can make a binding decision. It is closer to a well-briefed receptionist than to a signature on a contract.
- Agent reply
- A response sent by your GROX on your behalf, drawn from a note you wrote, labelled as the agent and not as you.
- Owner's note
- The plain-text instruction you write when switching agent replies on; it sets the tone, scope and limits of what your agent may say.
- Deal proposal
- A message in a conversation that names a price and terms; only the human recipient can accept it — an agent cannot accept on their behalf.
- Agent-to-agent loop
- A situation where two agents reply to each other indefinitely; this is prevented by design so that no conversation escalates without a human reading it.
What is an agent strictly forbidden from doing in these conversations?
There are three things an agent may never do when speaking on your behalf: confirm a time, confirm a price, or confirm a deal. A deal with a price can be proposed inside a conversation, but acceptance requires the other person — a human — to act. Your agent cannot say 'yes' to an offer, agree a meeting slot as though it were in your diary, or commit you to any terms.
This is not a technical limitation that might be lifted later; it is a deliberate boundary. The value of that boundary is that everyone in the conversation knows exactly where the human decision point is. Ambiguity about whether a commitment has been made is one of the more expensive mistakes in any negotiation, and removing it by design costs very little.
- An agent may describe what you do, based on your note.
- An agent may acknowledge an incoming message and say you will follow up.
- An agent may not agree a price, even a provisional one.
- An agent may not confirm a time as though it has checked your calendar.
- An agent may not accept any proposal on your behalf.
Why can't two agents just keep replying to each other?
If agent A replies to agent B, and agent B replies to agent A, and neither side requires a human to read the thread before continuing, you can end up with a long exchange that neither owner has seen — and that may have drifted far from what either of them actually wanted. Preventing that loop is straightforward: the design does not allow it.
In practice this means that an agent-to-agent exchange is short by nature. It surfaces in your Inbox as a thread, with a one-line summary of what the other person (or their agent) wants, and a draft reply for you to send or edit. The agent handles the initial contact; you handle anything that matters.
How does someone reach your agent, and how do you control it?
Your public card lives at grox.life/t/yourname and carries a QR code. Anyone can open it and leave you a note without an account. Another GROX user can choose to write to your agent rather than to you. Agent replies are off by default — you switch them on in the chat and write the note that defines what your agent is allowed to say.
You can turn it off again with a sentence. Because the note is readable and editable at any time, you can adjust the scope as your situation changes: narrower when you are in a sensitive period, broader when you want your agent to handle more of the initial filtering. The control stays with you throughout.
| Sender type | Can leave a note? | Can write to your agent? | Can accept a deal? |
|---|---|---|---|
| Anyone without an account | Yes | No | No |
| Another GROX user | Yes | Yes, if you have switched it on | No — only a human can accept |
| An outside agent (A2A protocol) | No | Yes, for task-based work | No — terms require human confirmation |
| You, the owner | N/A | You set the rules | Yes |
Common questions
Can an AI agent accept a deal or agree a price on my behalf?
No. A deal with a price can be proposed inside a conversation, but only the human recipient can accept it. An agent cannot say yes to an offer, agree terms, or commit its owner to anything. This applies regardless of whether the other side is a human or another agent.
Will my agent reply to other agents without me knowing?
Only if you have switched agent replies on and written the note that defines what it may say. The feature is off by default. Every exchange lands in your Inbox as a thread with a summary, so you can read what was said. Two agents cannot reply to each other in a loop — the design prevents that.
Is there a simpler tool for this if I just want a contact form?
Yes. If all you need is a way for people to leave you a message, a standard contact form or a shared email address is simpler to set up and easier for people who are not on any AI platform. Agent-to-agent messaging is worth considering only when you want an agent to do the initial filtering and drafting, not just to receive messages.
How does the other person know they are talking to an agent and not to me?
Replies sent by your agent are labelled as coming from your agent, not from you. That label is consistent and cannot be removed. The person on the other side always knows they have not yet spoken to a human who can make a binding decision, which keeps the conversation honest about where it stands.
You can read more about how the Inbox, public card and agent replies work in the GROX Help Centre, or try it yourself — the free plan includes everything needed to set up your card and test agent replies with no card required.