Lieben and Illingworth both post Claude Code cold email builds

Two operators published Claude Code workflows for cold email in the same week, and neither attached a send result to it.

Michel Lieben's version is a tool chain. He names six APIs that run the whole thing inside Claude Code: AI Ark to find companies, which he says holds over 70 million, PredictLeads to show which of those are hiring for the roles you care about, BlitzAPI to find the right people and their emails, LimaData to fill the emails BlitzAPI missed, FullEnrich for mobile numbers, and Instantly to turn the list into a campaign. His earlier post ends where most of these threads end, with the campaign sitting in Instantly ready to send.

Illingworth's version is seven steps, not six tools

Richard Illingworth opened with a verdict on the category before describing his own build. He starts by telling Claude what he sells and who buys it, and says it keeps asking follow-up questions until an ICP comes out the other side. He then hands it a handful of companies he already knows fit, and it builds Apollo search filters off their profile.

The tiering is the part worth noting. Tier 1 is anyone Apollo shows already running a sequencer, tier 2 is the wider pool held back for later batches. Enrichment scrapes every contact's website, job listings and recent news in parallel, and the output is a three-email sequence ending on a short breakup email.

The last two steps are the ones most people skip. Step six is a QC pass that checks every merge field has a column behind it and reads every single variant before launch. Step seven is a post-send read that splits reply rate and positive reply rate by segment, so he knows what to scale and what to cut.

What is actually transferable

Both posts are process descriptions. Neither says how many emails were sent, to whom, or what came back, so there is nothing here to benchmark against. What you can copy without taking anyone's word for anything is the structure: a merge field check that fails loudly before send, and a segment-level split of positive replies after it. Those two steps work with whatever stack you already run.

The tool list is the more fragile half. Six named APIs is a specific bet on six vendors, and Lieben's post does not say what any of them returned on a live campaign. Illingworth's build leans on Apollo filters and parallel scraping rather than a fixed vendor chain, which is easier to swap out when one source goes stale.

Richard Illingworth

@RCIllingworth

X

99% of AI personalisation with cold email is slop.

Michel Lieben

@MichLieben

X

the campaign lands in instantly, ready to send