Hermes: The AI Agent That Builds Lead Lists From Slack

The problem

He writes that every one of his agency’s campaigns “starts with a list.” He describes how that work used to look: “3 or 4 years ago, I was finding emails by hand, enriching lists from Apollo one by one, and uploading them to a verifier manually.” He now describes it this way: “Now it’s a Slack message.”

The blueprint (as he describes it)

Run the agent on an always-on server. Hermes lives on a VPS that stays on 24/7. “It never depends on someone’s laptop being open.”
One channel, one back end. Everyone messages the same agent in the same Slack channel, from a laptop or a phone. One Codex account does the execution, “so the whole team shares one back end.” He says his own part was “mostly the skill files it works from.”
Connect the data tools. Hermes is wired into somewhere between 20 and 40 databases and enrichment tools. These cover TAM sourcing, scraping, enrichment, email validation, account research and personalization. For standard lists, it pulls through Clay’s CLI.
Keep a skills repo. The team keeps a GitHub repo of written instructions for every task Hermes runs, “from how to scrape a source to how to qualify a lead.” Hermes reads those skills before every job. He says a good skill covers the source and how to reach it, the steps in order, the fields to pull, the checks that confirm a row is good, and what to do when a step fails. He says the last part matters most, and he advises writing 1 skill per job.
Write the brief in 5 parts. Every request names who the company is, who the person is, the custom data points written as yes/no checks the agent can verify, the output columns, and the size. His example asks for B2B SaaS companies with 10 to 100 employees, a VP of Sales, Head of Sales or other senior sales role, at least 3 SDRs or junior salespeople, and a live pricing page.
Answer questions, then check a sample. Hermes first asks clarifying questions, such as which people to target and the delivery format. It then returns a small sample. Each company comes with the employee estimate and its source (it pulls from multiple sources, including PitchBook), the live pricing page, the evidence it is B2B, the decision maker, and 3 salespeople with their LinkedIn profiles and titles. If the sample is right, it runs the full list. He writes that “a bad brief gets caught after 5 rows, before it hits 5,000.”
Scrape sources databases don’t cover. A client wanted construction companies that have built with mass timber. A public group lists those companies, and he says Clay can’t pull it. Hermes scraped the full source using the skills in the repo. He suggests industry association member lists, certification and award directories, public project databases, event exhibitor lists and government contract records.
Research the accounts. On the same example he asked for a 300 to 500 word research report on each company, and it came back with a full report on each account. Hermes delivered it as a separate file. He would rather have it as a column next to each account. He says this gets expensive at volume and is best for a small list of top-tier accounts.
Hand copy to Claude Code, then review. Hermes can write messaging, but the team uses Claude Code for copy because “working in the terminal is easier than going back and forth in Slack.” The full flow: Hermes builds and enriches the list, Hermes researches the accounts, Claude Code writes the messaging off the research, the campaign goes into the sending setup, and the team reviews everything before it sends.

Stated results

None stated for reply rates, meetings or cost for Hermes itself.
“The qualification rate on lists like this is much higher than anything we pulled the old way.” No number is given.
He says the campaigns “are performing great.” No metric is given.
He says Hermes “costs less to run than any database or workflow tool we’ve used, Clay included.” No figure is given.
Agency totals, not attributed to Hermes: “$70M+ in pipeline for 100+ B2B companies, off 10M+ cold emails.”

Caveats

He owns the agency. The issue ends with offers for GTM University, his paid community, and for booking a call with Pipeline.tech.
The agency totals are self-reported.
By his own account, his team built Hermes. He says his part was mostly the skill files.
The skills repo is private, so the instructions behind the results can’t be checked.
He calls the mass-timber example and the main prompt real, and says he ran the main one on camera. The ecom and recruiting requests are given as examples only, with no reported runs.
He lists tradeoffs: the agent goes down sometimes, APIs change, skills need updating, and it has to be built from scratch. He says that if GTM is only 1 part of your business, building and maintaining your own agent “is a lot.”
Handles: X @coldemailchris (verified 2026-11-03: bio reads “Founder/CEO @Pipeline_tech”). LinkedIn /in/coldemailchris (verified 2026-11-03). No Bluesky, Threads, Instagram or TikTok verified.

FAQ

What is Hermes?

Hermes is the AI GTM agent his team at Pipeline.tech built. It runs on a VPS that stays on 24/7, the team messages it in one Slack channel, and one Codex account does the execution.

Why does Hermes return a sample before the full list?

He says the sample step saves wasted credits, because “a bad brief gets caught after 5 rows, before it hits 5,000.” You check the sample, and if it’s right, Hermes runs the full list.

Can I see the skills Hermes uses?

No. The GitHub repo of skills is private, and he says the team will never show it publicly. He does describe what a good skill covers.

Did Hermes improve reply or meeting rates?

The issue doesn’t say. He gives no reply, meeting or cost numbers for Hermes, and says only that the qualification rate is much higher than the old way, without a number.

Sources

https://the-pipeline-newsletter.beehiiv.com/p/how-hermes-runs-our-entire-gtm-system (primary: first-person newsletter issue by Christian Plascencia, Oct 20, 2026)

Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).

5 Rows Before 5,000: How Christian Plascencia's team built Hermes, an AI agent that builds their lead lists