Blue MesaSummer 2026
A prospecting service that finds companies with a reason to buy now, built on parallel AI agents and one rule: every contact needs a source.
Two paying customers in summer 2026.
The problem
B2B sellers spend much of their time finding out which companies are worth contacting right now, and who to talk to there. Most companies in a market are not ready to buy at any given moment.
What I built
I designed and automated Blue Mesa's prospecting workflow with Claude Code. For each client, an onboarding questionnaire fills a configuration with what they sell and to whom. A pipeline of ten prompt templates then finds prospects, analyses them, finds a recent dated event that gives each one a reason to buy, builds contacts, runs a quality check, prioritises and writes outreach, and generates PDF reports.
For contact research I wrote a Claude Code skill that runs the work in parallel: the company list is split into groups of 15–25, one subagent per group, all started at once.
How
Claude Code with subagents and skills, Python (openpyxl) for reading and writing the spreadsheets, LibreOffice headless to check the output renders.
A source hierarchy instead of trust.
Each agent may only use an address from the company's own site, a press release or report where the person is listed, or a proven email pattern on the same domain. LinkedIn and company registries may confirm a role, never an email address.
Every row is classified.
Verified (with link), pattern (with the example address that proves the format), generic, or not found. Not found is a valid answer.
Code writes the result, not the model.
Agents save JSON; a script writes it into the spreadsheet, never overwrites old data silently, and a check script compares the new file with the original so no rows or values go missing.
The hardest part
Stopping a language model from inventing plausible email addresses. The answer was to make “not found” cheap and fabrication impossible to hide: every address must carry its source and class, and the colour coding in the delivered file shows the client exactly how confident each row is.