Reed, the Reply Agent: intent and how replies are written

Reed is our named, ready-to-use Reply AI Agent template. Perfect for beginners and easy to tune.

Written By Kevin Lawrie

Last updated About 1 month ago

Reed is our named, ready-to-use Reply AI Agent template. Where basic reply sentiment tells you a reply is broadly Interested or Not Interested, Reed classifies it into a much more specific intent, then drafts a reply built for that exact situation.

How Reed works: classify, then act

Reed runs in two stages every time a lead replies.

Stage 1: Classify. Reed reads the inbound reply alongside the full outbound history of that campaign thread, and assigns it exactly one intent token from a fixed list.

Stage 2: Act. Based on which token was assigned, a matching conditional action runs a campaign reply, generating a short, tailored message using an [[AI:...]] block written specifically for that intent.

This is a meaningfully different approach than a single broad sentiment label. A reply asking "can we set up a call" and a reply saying "we're already using a competitor" are both technically positive-ish signals a coarse classifier might lump together, but they call for completely different responses. Reed tells them apart and responds accordingly.

Reed classifies the first inbound reply in a campaign thread. That's the moment where getting the response right matters most.

The intent tokens

Reed sits inside the broader sentiment bands (Interested, Not Interested, Re-engage, Wrong person) but picks a specific token within them.

Positive and open intents

Token	What it means
`meeting_request`	A clear yes to a call, a proposed time, or acceptance of a meeting ask
`info_request`	Wants more detail, a demo, or an example before committing to anything
`referral_redirect`	Interested, but pointing you toward someone else at their company
`timing_later`	Interested, but wants to revisit at a specific later point
`positive_acknowledge`	Warm tone, but no clear next step offered

Objection and wrong-person intents

Token	What it means
`objection_timing`	Bad timing or too busy right now, not a hard no
`objection_budget`	A price or budget concern
`objection_incumbent`	Already using a competitor
`objection_relevance`	Doesn't see how this applies to them
`objection_authority`	Not their decision to make, prefer `referral_redirect` if they've pointed you elsewhere
`unclear`	Reed can't confidently classify this one

What each intent's reply actually does

Intent	Reply behavior
meeting_request	Thanks them and asks for the best day or time, or confirms a time they've already proposed
info_request	Answers what they asked, includes one concrete detail from your own product or ICP, closes with a soft next step
referral_redirect	Thanks them for the redirect, acknowledges any named person or Cc'd contact directly, only asks for an intro or email if that contact path is still missing
timing_later	Acknowledges their timeline, asks when it makes sense to follow back up
positive_acknowledge	Acknowledges the tone, offers one low-friction next step
objection_timing	Brief acknowledgment, asks when would be better
objection_budget	Reframes value without discounting, offers to walk through the numbers
objection_incumbent	Doesn't disparage the competitor, asks whether their current tool covers the specific capability this campaign is about
objection_relevance	Doesn't argue the point, offers a gentler reframe for their specific role, leaves an easy out
objection_authority	Acknowledges any named or Cc'd contact directly, or asks who the right person is if none was given
unclear	Acknowledges the reply, asks one clarifying question

Identity rules

These matter in every reply Reed writes, and they're worth understanding regardless of which agent you use.

The prospect is whoever replied, identified only by their name, title, and employer, plus whatever they've said in the campaign thread and this reply. Reed does not invent facts about them beyond that.

{{firmographics}} is your workspace's product and ICP, never a description of the prospect's company. Reed is built to never confuse the two.

Replies stay short, direct, and human. No corporate filler, no long paragraphs where a sentence will do.

Referrals and Cc'd contacts

When a reply names a referral or already has someone on Cc, Reed acknowledges that person by name rather than asking a generic "who should I loop in?" question. If prior campaign messages give Reed enough context, it can offer a brief, relevant line to that new contact rather than repeating a full pitch. It only asks for an introduction or an email address when that contact path is genuinely missing.

Customizing Reed, or building your own

Reed works well out of the box, but you can clone it and adjust its behavior, or build an entirely custom Reply Agent from scratch using the same underlying structure. Two levers matter most for getting good output:

Your workspace's Firmographics, which shapes how it talks about your product
Extra guidance added directly into the classifier prompt or the per-intent reply templates