Christian's team wired a Slack agent that builds their TAM
One shared Codex account, a Hostinger VPS and a GitHub repo of GTM skills sit behind the list build.
Christian published the full build for an agent his team uses to assemble their TAM (total addressable market) from a single Slack message. You type what you need in the channel and it comes back with verified contacts. He credits the build to @YourVibeGuy and @shaw_gtm and says he takes none of the credit himself.
The setup
The agent is Hermes, from Nous Research, running on a Hostinger VPS. Underneath it sits one Codex account rather than Claude Code, and Christian gives two reasons for that choice, cost and stability across large jobs.
The VPS keeps it running 24/7 inside the team Slack, so everyone works off that single subscription. Christian puts the saving at about $800 a month against paying for a Codex seat per person. Each person also gets their own Slack channel with the agent, which he says stops one person's list build bleeding into someone else's context.
What it can call
Every tool is connected by API, MCP or CLI so the agent can call any of them mid-task. The list Christian published runs to sixteen names, including Clay, AI Ark, Icypeas, Prospeo, LeadMagic, MillionVerifier, BounceBan, Firecrawl, Serper, Exa, Ocean.io, Spider Cloud, GetLeads, Apify, GTME.Business and OpenWeb Ninja. That is three enrichment providers and two verifiers sitting side by side, plus several scrapers, which is what you would expect from a team that wants a fallback on every waterfall step rather than one vendor per job.
The part that is hard to copy
The tool list is the easy half. Hermes also has access to the team's GitHub repo, which Christian says holds a skill for every part of GTM they run. Ask it for anything and it pulls the skill for that job before it starts, so the output follows playbooks the team has already written rather than whatever the model decides to do.
That is the difference between this and a generic agent with API keys. Without the repo, the sixteen integrations are just credentials. With it, the agent reproduces house process. Christian shared a video of a list being built from one Slack message and noted the live example starts at 3:10.
For people sending, this is a continuation of a pattern already visible in this feed, teams moving list building out of point tools and into coding agents. Lieben earlier described running campaigns in Claude Code with no Clay at all. Christian's version keeps Clay in the stack but puts an agent in front of it, and the reported per-seat saving is the concrete claim here, not a reply rate.
We picked Codex over Claude Code because it's cheaper & holds up better when a workflow runs across thousands of rows.
Ask it for anything & it pulls the skill for that job before it starts.