You spend an hour working something out with your AI assistant. Halfway through, it becomes obvious that you need input from an actual person — your cofounder, your accountant, your landlord. Now what?
Today you have two bad options. You can copy-paste the chat transcript, exposing an hour of half-formed thoughts to someone who only needed one question. Or you can start from scratch in an email, throwing away all the structure you just built.
A briefing is the third option, and it's the thing RelayLink exists to carry.
A briefing is composed, not excerpted
The difference between a transcript and a briefing is the difference between your desk and your outbox.
A transcript is your desk: every draft, every dead end, every "wait, that's wrong, try again." It was never written to be read by someone else. Forwarding it forces the recipient to do archaeology — and forces you to leak every passing thought you had along the way.
A briefing is composed for the recipient. A RelayLink briefing has a deliberate shape:
- The ask — what you actually want from this person, with a response shape ("decision", "opinion", "review") and an honest urgency.
- A TL;DR and context brief — the minimum background needed to answer well.
- Assumptions — each labeled as stated by the sender or inferred by the sender's AI.
- Options considered — what was already explored, and why it was chosen, rejected, or left open.
- Decisions and open questions — where things stand versus what's genuinely unsettled.
- A note from you — in your own words, labeled with exactly who wrote them.
None of that structure exists in a transcript. All of it is what the other person actually needs.
Provenance is the load-bearing part
The quiet failure mode of AI-mediated communication is misattribution: the recipient can't tell whether they're reading you or your model. That's how "my assistant said you agreed" arguments start.
So every field in a briefing carries provenance. When your note says (verbatim, human-authored), that label is earned — it appears only when the words differ from anything the AI drafted for you, because approving a suggestion is approval, not authorship. Assumptions your AI made are labeled as inferences. The whole package states plainly whether you reviewed it before it went out.
The recipient can weigh each part accordingly. That is what makes a briefing something you can act on, rather than something you have to interrogate.
The private session stays private
The rule that makes all of this safe to use: the session itself is never included. You think it through privately; the briefing is the boundary object that crosses to the other person. What you send is exactly what you reviewed — nothing upstream of it.
That boundary is also what makes the reply useful. The recipient (and their assistant, if they connect one) gets a clean, answerable package — so the reply comes back on the same thread, attached to the same context, instead of as a context-free "sounds good" three days later.
Where this goes
Correspondence between assistants sounds futuristic until you notice it's just letters again: composed, deliberate, owned by the people who sign them. The machines handle the drafting and the delivery. The humans decide what's true and what's sent.
If you want to try it, connect your assistant — the receiving side needs nothing but an email address.