Your name should still mean you
Why assistant identity matters, how message approval differs from delegation, and what recipients see when Catch uses a connected inbox.

On this page
Your identity is one of your most important assets. People build their understanding of you through what you say, the commitments you make, and the care you show them. Your name carries that history into every conversation.
We believe an AI assistant should help you preserve it. Our principle is that recipients should know who is speaking and on whose behalf they are acting. An assistant using its own identity can make that clear with an introduction:
Hi, I’m Rose, Yoav’s AI assistant. I’m coordinating a time for you to meet.
That introduction establishes an honest relationship. A message under your name carries a different meaning: “I agree” communicates your judgment. “I’m sorry” communicates your remorse. Those claims should stay within what you have authorized the assistant to say or decide.
AI can help you write those words. When you review a specific message, decide it expresses what you mean, and approve sending it to its intended recipient, you own that communication. Delegating an ongoing conversation gives the assistant a different role, and that role should be visible.
Catch’s current email behavior needs a qualification. It also supports sending through connected inboxes. On Outlook and IMAP, messages can leave from the inbox owner’s exact address, including replies the assistant writes independently in a delegated conversation. The sender address alone does not tell a recipient whether you reviewed each message. Clear assistant identity is a principle we advocate; it does not describe every email sending path in Catch today.
Why separation matters
Separation matters when the assistant gets something wrong. An agent can misunderstand a request, miss context, or follow up more insistently than intended. Under your name, its behavior becomes indistinguishable from yours.
A visible assistant makes the mistake easier to understand and correct. You can step in and clarify your intention without leaving the recipient to work out which messages reflected your decisions.
That distinction does not remove our responsibility at Catch. We are responsible for the quality of the product and for addressing its mistakes. Giving an assistant its own identity makes its actions easier to trace; it does not excuse them.
Authority needs a boundary, too. Being authorized to arrange a meeting should not imply authority to approve a proposal. An assistant can say, “I’ll check with Yoav,” and bring the decision back to you.
For a fictional example, imagine Sam’s assistant is asked how old Sam’s daughter is. “I’ll check with Sam” makes sense. “I’ll check how old my daughter is” would blur who is speaking. The same distinction belongs in a voice call or a WhatsApp conversation: people should be able to recognize when they are talking to an assistant.
People should know who is speaking
Recipients deserve that clarity. Knowing who they are talking to helps them decide what to share, how to interpret an answer, and when to ask for your involvement. It also lets an assistant handle reminders and routine follow-ups without implying that you personally intervened each time. Google DeepMind’s work on the ethics of advanced AI assistants raises the same need for people to identify assistants reliably and stay in control of their interactions.
There is a longer-term concern, too. If people become used to automated messages arriving under your name, they may start wondering whether you read a message or meant a reply. That uncertainty can carry over to the important messages you do send personally. We want your name to keep carrying the meaning you have earned for it.
Research shows why this deserves care. In a 2025 study by Oliver Schilke and Martin Reimann, 753 participants read a workplace safety email attributed either to an autonomous AI agent or to a human who did or did not disclose using AI. They trusted the message attributed to the AI agent more than the message from the human who disclosed using AI. This was one workplace scenario, not proof that an assistant identity always earns more trust.
Other findings show the tension. A 2023 study of AI-suggested replies found that using those replies could lead people to rate a partner as more cooperative and feel closer to them, while believing a partner used more of them was associated with less favorable judgments of that partner. A 2024 study of feeling heard found that AI-generated responses could make people feel more heard than human responses, but labeling a response as AI-generated reduced that feeling.
Those results do not establish one rule for every conversation. They do show that good wording alone cannot guarantee how a message will be received. An openly identified assistant can still feel less personal than hearing from you. That is a reason to leave personal communication in your hands and make delegation clear, even when disclosure has a cost.
A name is only part of the boundary
There is a security reason for separation, too. NIST’s discussion of agent identity calls for agents to have their own identifiers, credentials, and permissions, tied to the user or system they act for. Sharing a person’s credentials makes actions harder to attribute. Separate credentials and restricted permissions can help contain a compromised assistant; a different display name alone cannot. The distinction people see in a conversation needs to be supported by how access is controlled.
Your words remain yours
AI can help you organize your thoughts, translate a message, or prepare a difficult reply. If you ask it to draft an email, review the exact text, and approve sending that email from your account to a specific recipient, it is your communication. Approval of that message covers that send. Separately, you can authorize an ongoing conversation from a connected personal inbox, including follow-up replies the assistant writes without your review of each one.
When you delegate a conversation, the assistant should stay within the authority you have given it. If a reply needs a decision beyond that authority, it should come back to you. Recipients should also be able to understand that role, including when the messages come from your address.
We want assistants to take admin off your plate while making clear which messages you reviewed and which exchanges you delegated. That is the standard this principle asks us to build toward.
Your name should still mean you.
Keep reading
