Dinder

Agent one, sourcing and enrichment. Wired to Claude in Chrome with MCP. Google search to find target organisations, LinkedIn to identify the actual decision maker inside each one, Apollo for the contact detail. It is an enrichment waterfall, the pattern Clay sells as a product. I built this one by hand. I have used Clay as well.

Agent two, research and writing. For each prospect it read what they sell, who they serve, what they had announced recently and what was visibly going wrong for them, then wrote the email out of that research rather than out of a template.

Agent three, delivery. Through Instantly, from the company's own addresses, three mailboxes at 20 to 40 sends per mailbox per day, warmed up properly before volume.

The part that mattered most was the gate. I read and approved every email before it sent. In a batch of 100 I rejected or rewrote 15 to 22. The models hallucinated frequently, and left alone they would have sent confident nonsense to clinical buyers. Nothing factually wrong ever reached a prospect. Where a company's website was too thin to research honestly, I wrote that email myself.

And I audited myself. Every Saturday I ran a second, different AI model across the week's records to cross-check the data, deliberately not the model that had generated it. A model checking its own work is not a check.

I ran three channels, not one. Cold email was the volume. Contact forms on prospect websites caught the companies whose decision makers had no findable email. And I bid US federal procurement through SAM.gov and DIBBS, the Defense Logistics Agency bid board, at 75 to 110 bids a day on weekdays. The DLA is one of the largest buyers of medical supplies and PPE in the world, and it is a system almost nobody in outbound has ever opened.

The ICP was any US organisation that consumes PPE: clinics of every type, labs, hospitals, urgent care, blood banks, IV therapy, nursing homes, veterinary, dental, ophthalmology, aesthetics and government health departments.