Illingworth adds QC pass and segment cuts to Claude Code stack
Two steps past the version he published last week, with a caveat about inboxes attached.
Richard Illingworth reposted his Claude Code cold email workflow with two steps beyond the version he published last week. Step five is a QC pass Claude runs on every variant before anything gets sent. Step six is a read on the data once it exists, telling him which segments to scale and which to cut.
The rest of the stack is the same shape. He gives Claude the ICP and the offer, and it pulls a targeted list from Apollo. It enriches every contact in parallel, scraping their site and pulling their tech stack. It writes the sequence off that enrichment, so each email references the prospect's real setup. He drops the output into Instantly, maps the merge fields, caps the daily volume and turns off open tracking.
What the two new steps actually do
The QC pass is the interesting addition, because a generated sequence off scraped enrichment is exactly where an AI build breaks. Variants that reference the wrong tech stack or a site that no longer says what the scrape said go out at volume before anyone reads them. Putting a check between the writing step and the send step is the part of the workflow that a human used to do by eye.
The segment call at the end is the other half of the same problem. Illingworth does not publish reply rates, volumes or which segments got cut, so there is no number attached to either step. What he is describing is a process, not a result.
The caveat is the part to copy
Illingworth closes the post with a line in brackets that does more work than the six steps above it. He says none of it works without real inboxes underneath. That is the honest read on every AI-run outbound build going around right now. List pulling, enrichment, sequence writing and QC all sit on top of the sending setup, and none of them repair a weak one.
For people sending, treat this as a build order rather than a shortcut. Illingworth says he has sent millions of cold emails over the last couple of years, and Claude Code is running a process he already knew how to run by hand. The parts he kept manual are telling. He maps the merge fields himself, he caps the daily volume himself, and he turns off open tracking himself. Those are the decisions that decide whether the infrastructure survives the campaign, and they are the ones he did not hand over.
There are no numbers in the post. Anyone copying the stack is copying a workflow one operator says he uses, not a tested lift in replies.
(none of this works without real inboxes underneath)
once there's data, it tells me which segments to scale and which to cut