How I design sales agents, what I've built, what I'm testing now, and the prompts, workflows and skills I use. Take what's useful. Then hold it to the 95% rule before you trust it.
I cloned my voice with ElevenLabs and built an assistant trained on this website. Ask it about your sales engine, how I work or what I've built. It tells you straight away it's an AI, and I read every conversation myself.
AI can do a lot of the work. It can't replace the people at the top who understand the market, the culture, the customer and the nuance in a message. Strategy stays human. The agents carry it out, and your team spends the time it gets back with buyers, where trust is won.
Nobody expects a new rep to be great on day one. Agents are the same: give them your best examples, correct them, talk to them, and they get better. Skip that, and you'll never get the output you need.
Treat AI like a lean start-up. Small tests, real data, quick changes, and nothing scaled until it works. It's a constant learning process, not a one-off install.
Not one big bot. A set of small specialist agents, each easy to test, with people approving until the work earns the right to run alone. Tap any part to see how it works.
Built with HubSpot and Claude Code, with Apollo, Lusha and LinkedIn Sales Navigator for enrichment. Several agents in one flow: one enriching every lead, one scoring fit from the profile, one writing the message. Each new lead landed in a Slack channel with the qualification, the details that mattered and a reply already drafted. The team could check it, give feedback right there on the message, and approve to send, so every piece of feedback was collected in one place.
First reply went from about a day to 5 minutes.
Clay and Claude Code, using app store data to find companies where the product could make a big difference, with Lemlist warming up the sending addresses. Then a lot of testing on the message by person, company and timing.
Largely automated, hitting about 500 accounts and 4,000 contacts a month.
The scorecard on this site sends a report from my own inbox through a small Google Apps Script, records the lead in HubSpot and offers a call through Cal.com, which logs the meeting too. It cost nothing to run, and the report lands in the inbox, not spam.
The full build is in the free Workflows pack below.
ElevenLabs for the cloned voice and the conversation, trained on this website, with a clear AI notice up front. After each conversation it sends me a summary, and the visitor goes into HubSpot only if they agree.
A daily scan of funding news and the best go-to-market writing, an idea list with my take, and drafts written from a profile of how I speak and write. I approve every word before it goes out.
The test is simple: would your best person send this, as it is? Only when the answer is yes 95% of the time does an agent stop needing a person to press approve.
I'm learning in public. Each of these gets written up in the newsletter: what worked, what didn't, and what I'd do next.
Scans X, a set of Substacks and funding news every day, and gives me the highlights worth acting on. The best ideas become builds, and the builds become posts.
Starting with ElevenLabs: where a voice agent could help a sales team, and where a person on the phone is still the point.
Testing HeyGen for short videos alongside clips I film myself, to see what's worth automating and what isn't.
A personal CRM for my own relationships, so I follow up and reply to everyone, and a rituals and habits app that works on my phone and the web. Small tools, built to keep my hands on the tech.
Each prompt is built for production: a role, real context, clear steps, rules, and output another agent can read. The first is here in full. Get all seven, with what to feed each one, how to check it and where it goes wrong, in the free Prompt Pack.
The full prompt, checks and failure modes are in the free Prompt Pack.
Three PDFs, mapped to the layers of the Coburn Commercial Engine. Free. I'll email you the link, and you can open it straight away.
Seven production prompts, one for each layer of the engine, with what to feed each one, how to check it and where it usually goes wrong.
Three builds step by step: the inbound agent, the outbound signal engine and a lead engine on free tools. Every agent, every check, how to roll it out and what to measure.
Five skills that teach Claude to do sales jobs your way: account research, call prep, follow-ups in your voice, pipeline review and ICP fit, with the full SKILL.md files.
In a Commercial Systems Build I set up agents like these on your own deals, calls and CRM, and only switch them on once they send what your best people would send.
One lever or one AI workflow per issue, what I'd change this month, and what happened when real companies did it. Short, practical, no hype.
Apply in one minute. If it's a fit, you book a call straight away and leave with at least one thing to fix this month, whether we work together or not.