The problem this solves
Large list builds fail expensively when the criteria were slightly wrong. Five hundred rows built on a misunderstanding of 'mid-market' or on the wrong reading of a role title cost the credits, the review time and the confidence of the person who asked, and the correction means running it all again. The fix is old and simple: show a sample first, correct it in conversation, then run the rest.
Doing this in a thread keeps the conversation and the work in one place. The criteria are typed as a message; ten rows come back under it; the corrections - 'drop agencies', 'VP or above only', 'add Canada' - are typed as replies; the next ten rows reflect them; and only when the sample looks right does the full build run. The thread becomes the record of how the list was defined, which matters when someone asks six months later why a company is on it.
The Prospecting Agent runs the whole loop from the thread, and the final push to HubSpot is a reviewed batch, so the CRM receives only contacts that passed both the sample review and the final look.
How the mission runs
- Read the criteria from the thread. The Prospecting Agent reads the thread for the criteria - industry, headcount, region, technologies, roles, exclusions - restates them as a structured definition in one message, and asks about anything ambiguous before spending anything.
- Build and post ten sample rows. Ten companies and contacts matching the definition are built with the standard columns - domain, headcount, industry, the technology or signal evidence, the contact with title and verification status - and posted in the thread as a compact table.
- Take corrections in the thread. Replies in the thread adjust the definition; the agent restates the changed definition and posts a fresh sample of ten. The loop continues until you reply that the sample is right, and each version of the definition is kept.
- Build the full 500 and save the list. The full build runs against the agreed definition, with verification on every contact, and the result is saved as list <name> with the definition attached, so it can be refreshed later with the same criteria.
- Push approved contacts to HubSpot. The contacts are staged as a HubSpot batch with counts of creates and updates posted in the thread; on your reply approving it, the batch is written with the list name as the source, and a final summary closes the thread.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
List <name> with 500 companies and verified contacts built to a definition agreed in the thread through sample rounds, saved with the definition attached, and the approved contacts written to HubSpot as a reviewed batch. The thread holds the definition's history, the samples, the corrections and the final counts, which is the audit trail for the list.
Make it yours
- Ask for 20 sample rows or two samples from different segments when the definition is broad.
- Push to Pipedrive, Attio or a Google Sheet instead of HubSpot.
- Skip the CRM push and keep the list in the workspace for scoring and audiences first.
- Run the same loop in Telegram or by email for people who do not use Slack.
Frequently asked questions
How many sample rounds are typical?
Two or three. The first sample usually reveals one misunderstanding, the second confirms the fix, and the full build follows. Each round costs only the ten rows it produces.
What if the definition matches fewer than 500 companies?
The agent says so in the thread with the true count and proposes which criterion to relax; you decide whether to widen or accept the smaller list.
Can the definition be reused?
Yes. It is saved with the list, so a refresh or a similar list for another region starts from it rather than from scratch.