Lieben builds a campaign from one prompt in 15 minutes
Ten tools, 50 CEOs, 48 enriched contacts and 13 emails that landed as drafts while he was away from the desk.
Michel Lieben says an entire cold email campaign can be run from a single prompt in about 15 minutes, using Claude Code with ten tools connected to it. He tested it on the 50 CEOs of the Forbes AI 50. The run produced 13 emails, loaded as drafts, while he was on a break.
What the machine actually does
Lieben lays out the chain in five steps. A database builds the list, with Prospeo and Apollo named. An enrichment tool finds emails and mobile numbers, and he says FullEnrich returned 48 of the 50. A signal tool checks who raised money and who is hiring sales reps, which he credits to PredictLeads. A copy skill writes one email per CEO, which he attributes to ColdIQ. A sender then loads the campaign as a draft in Lemlist or Instantly. He says every prompt is published in the article he linked.
That is seven tools named against the ten he says are connected, so the rest of the stack is not spelled out in the post.
The number to ask about
The funnel inside the run is the part worth reading twice. Fifty CEOs in, 48 contacts enriched, 13 emails written. Lieben does not explain the drop from 48 to 13, and the obvious candidate is the signal filter, since money raised and sales hiring are gating conditions that most of a 50 company list will fail on any given week. Either way, the headline claim is that a prompt produced 13 drafts, not 48 sends.
For people sending, this is a demo on a famous 50 company list, not a campaign with reply data attached. Nobody has seen what the ColdIQ copy skill returns at volume, and a run that writes 13 emails in 15 minutes tells you nothing about what those 13 emails do once they land. The useful signal is the architecture, a list step, an enrichment step, a signal step, a copy step and a sender step, each handed to a tool and orchestrated from one prompt.
It is also the third public build of this shape in a week. Lieben earlier described a Claude Code campaign that took 27 minutes to build and loaded into Instantly. The pattern across these posts is operators moving the glue work out of Clay style interfaces and into an agent they can prompt, then measuring the build in minutes rather than the output in replies. Until someone posts send volume and reply rate from one of these runs, the time to build is the only number on offer.
the run wrote 13 emails and loaded them while i was on a break