Engineering July 2, 2026 · 2 min read
Putting an AI BDR Into Production: Meet Boston
TL;DR
Boston is an AI BDR running in production behind my contact form. Rather than bolt a chatbot on, I researched what makes a great BDR, talked to sales managers about how they qualify, turned that into a technical spec, and built the system with Claude. Every inbound lead gets researched, legitimacy-checked, profiled, matched to services, and answered with a drafted first reply, all queued for my review before anything sends.
This week I pushed the first AI agent into production to help run the business, an AI BDR (business development rep). I’m calling it Boston.
The interesting part isn’t that it uses AI. It’s how it was built: like a system, from a spec, with a human in the loop by design.
Start with the role, not the tool
Before writing any code, I wanted to understand what actually makes a great BDR. I researched the role and its responsibilities, and what separates high-performing BDRs at SaaS companies from everyone else. Then I talked to a few sales managers about how they qualify opportunities, prepare for conversations, and decide what deserves their attention, and how they define a qualified lead.
That’s the step most “AI agent” projects skip. If you don’t understand the job, you automate a caricature of it. The research is what makes the agent useful instead of just fast.
Turn research into a spec
I turned that understanding into a technical specification, then worked with Claude to design and build the system against it. The spec is the point: it’s what you build against, review against, and improve against. Vibes don’t survive contact with production; a written spec does. (This is the same engineering discipline I’d bring to a client’s marketing system.)
What Boston does
Now, every time someone submits a form on the site, Boston goes to work. It:
- stores the lead in an in-house CRM,
- researches the company,
- evaluates whether the inquiry appears legitimate,
- identifies likely business challenges and builds a business profile,
- recommends which services are most relevant,
- drafts a personalized first response with thoughtful qualifying questions,
- and prepares everything for my review before a single email is sent.
That last clause is the whole philosophy: human-in-the-loop. Boston gets the work to a reviewable draft; a person decides what actually goes out.
0 to 80%, not 0 to 100
The goal here isn’t to replace judgment; it’s to automate the busywork and get from 0 to 80% as fast as possible, so my attention goes to the 20% that moves the needle: the conversation, the judgment call, the relationship. AI is very good at the research, the drafting, and the preparation. It should do that, and hand a human the part only a human should own.
If your organization is figuring out where AI agents fit into real workflows, that boundary (automate the preparation, keep the judgment) is the one worth getting right. Curious what Boston drafts for a real inquiry? Start a conversation.
References & further reading
- Infrastructure & Systems, how AI-enabled workflows create leverage in a marketing system.
- Anthropic, Building effective agents, patterns for reliable, reviewable agent systems.
Key takeaways
- Start with the role, not the tool: understand what a great BDR actually does before automating it.
- Turn domain research into a written technical spec: that spec, not vibes, is what you build against.
- Human-in-the-loop by design: Boston drafts and prepares; a person approves before a single email sends.
- The goal isn't 0 to 100; it's 0 to 80% automatically, so your attention goes to what actually moves the needle.
Want this applied to your marketing website?
Book a call →Get the weekly newsletter
One signed take on brand, content, systems, and outcomes.