← All agents
Prospecting

comment-to-dm-agent

People commenting on your posts are the warmest list you are not working. This agent works it, one approval at a time.

Watches your LinkedIn posts, qualifies the people who comment, files them in your CRM, and stages the DM. A human approves every send.

What it does

A lead-magnet post generates a burst of comments, and the window to respond is short: a commenter who hears back the same day converts very differently from one who hears back next week. Working that list by hand means re-reading every thread, checking who was already contacted, and copy-pasting profile details into the CRM.

This agent polls the posts you register, classifies each comment's intent, and turns genuinely interested commenters into CRM records and Slack cards. Each card carries the person's profile, their company when it can be resolved, and a one-click send for the lead-magnet DM.

The doctrine on every send: Pierce proposes, you approve. No message leaves without a human clicking the button, and the agent never cold-messages beyond your first-degree network.

Before and after

Before

  • Comments pile up while the post is hot, replies go out when it is cold

  • Someone re-reads every thread to find the real prospects

  • Who was already contacted lives in someone's memory

  • Commenters never make it into the CRM

With the agent

  • Interested commenters become CRM records and Slack cards within the hour

  • Intent classification separates prospects from congratulations

  • A permanent ledger guarantees nobody is contacted twice

  • Every DM is staged, and a human approves every send

How it works

Seven gates. Most of them exist to make sure the agent stays polite: dedup, network distance, daily caps, and a human on the send button.

  1. 1

    Watch only the posts you registered

    The agent polls a registry of posts you explicitly enrolled. It does not crawl your whole feed or anyone else's.

  2. 2

    Classify each comment's intent

    Comments are classified as interested, not interested, or ambiguous. Only genuine interest becomes a card; congratulations and tags do not.

  3. 3

    Dedup ruthlessly

    A permanent seen-ledger, an already-contacted check against the CRM, and an open-conversation check. If a conversation already exists, the card warns instead of risking a double-send.

  4. 4

    Check network distance

    DMs are staged only for first-degree connections. Beyond first degree the agent does not reach out cold, full stop.

  5. 5

    Create the CRM record first

    No card without a person record. If the commenter is missing from the CRM, the agent creates the record from their profile before anything else happens.

  6. 6

    Propose the DM in Slack

    The card shows who they are, what they said, and the staged message. Pierce proposes, you approve: one click sends, one click dismisses.

  7. 7

    Respect human limits

    Daily caps on cards and outreach, and an active-hours window so nothing fires at 3am. The agent is built to look like a considerate human, because a human approved everything it sent.

Tools in this agent

Unipile

LinkedIn access: comments, profiles, network distance, and the DM itself.

Visit →

Attio

The CRM of record. Any CRM with an API fits the same slot.

Visit →

Slack

The approval surface: one card per prospect, one click per send.

Visit →

Some tool links are referral links. They never change what the tool costs you.

Built and run in production by Pierre Nicolas. This page describes a workflow that actually runs, not a concept. Last reviewed 2026-07-11.

The detection and reporting methods behind these agents are published as open-source skills and on GitHub.

Related agents

Pierce runs this for you

Every agent on these pages runs on real accounts, in Slack, with every write approved by you. Bring your stack; Pierce brings the discipline.

Pricing lives on the main page.