@milesgurule-png4Founder, Forge Labs
I built an AI sales team for my web agency. Here's what the data taught me.
I'm 15 and I sell $2,000 websites to local businesses by walking in with a mockup. This month I automated everything except the walking:
- Leads: 5,300 businesses from OpenStreetMap and the chamber directory, every website audited for problems an owner can check on their own phone.
- Mockups: Claude Opus designs a custom site from the business's real photos, critiques its own screenshots, and rebuilds. About 15 minutes each.
- The team: 4 Claude Code agents (director, lead scout, mockup maker, outreach writer) write me a brief each morning: who to visit, when, and what to say.
What surprised me, all measured:
- 15% of "your website is down" leads loaded fine the next day. Now a site has to fail on two different days.
- Half the "no website" businesses actually had one. Now that's a question I ask at the counter, not a claim.
- Small SEO issues were stacking into top scores. Polish problems alone now can't make a lead "hot".
- Running the mockup build at high effort made Opus think for 25 minutes and never write the page. Low effort plus a separate critique pass worked.
The rule I'm proudest of: the agents can't send anything. They only write drafts; I'm the one who walks in.
Builders doing local sales: what's your best walk-in opener?
Comments4
My opener would be handing the owner the mockup built from their own photos and asking, "Is this close to how you'd want customers to see you?" Leading with their business looking good, instead of pointing out what's broken on their current site, keeps them from getting defensive.
The strongest part of this build is how you let the findings change the rules. Rechecking outages, treating “no website” as a question, and keeping polish issues from inflating lead scores all protect your credibility at the counter.
Keeping outreach in your hands adds accountability where it matters.
An opener I’d test: “Hi, I build websites for local businesses. I put together a quick concept for yours. Would you be open to a 30-second look?”
Lead with something useful, then let the owner tell you which problems actually matter.
Great hustle for the micro-SMB segment! But there’s a crucial architectural boundary here: this multi-agent cold-outreach model only scales at the lower end of the market (local service businesses). For Enterprise / Key Account Management, this approach completely collapses. Enterprise sales isn't a volume game of finding broken links; it requires deep contextual trust, navigating complex DMUs, and strategic alignment that standard LLM scrapers simply can't capture. The real question for agentic sales isn't just automating volume, but knowing where the cognitive cutoff lies between transactional automation and high-stakes B2B relationship building.
To add to this from a pure unit-economics perspective: Running a 4-agent stack with Opus for continuous vision critiques and scraping incurs significant token burn and maintenance drift as web layouts/APIs change. Beyond a certain scale, the TCO (API costs + maintenance hours) often exceeds the cost of simply buying a verified B2B lead database. Automating the workflow is impressive, but keeping the software stack stable can easily eat the profit margin if you aren't factoring in long-term maintenance overhead.
4 comments
My opener would be handing the owner the mockup built from their own photos and asking, "Is this close to how you'd want customers to see you?" Leading with their business looking good, instead of pointing out what's broken on their current site, keeps them from getting defensive.
The strongest part of this build is how you let the findings change the rules. Rechecking outages, treating “no website” as a question, and keeping polish issues from inflating lead scores all protect your credibility at the counter.
Keeping outreach in your hands adds accountability where it matters.
An opener I’d test: “Hi, I build websites for local businesses. I put together a quick concept for yours. Would you be open to a 30-second look?”
Lead with something useful, then let the owner tell you which problems actually matter.
Great hustle for the micro-SMB segment! But there’s a crucial architectural boundary here: this multi-agent cold-outreach model only scales at the lower end of the market (local service businesses). For Enterprise / Key Account Management, this approach completely collapses. Enterprise sales isn't a volume game of finding broken links; it requires deep contextual trust, navigating complex DMUs, and strategic alignment that standard LLM scrapers simply can't capture. The real question for agentic sales isn't just automating volume, but knowing where the cognitive cutoff lies between transactional automation and high-stakes B2B relationship building.
To add to this from a pure unit-economics perspective: Running a 4-agent stack with Opus for continuous vision critiques and scraping incurs significant token burn and maintenance drift as web layouts/APIs change. Beyond a certain scale, the TCO (API costs + maintenance hours) often exceeds the cost of simply buying a verified B2B lead database. Automating the workflow is impressive, but keeping the software stack stable can easily eat the profit margin if you aren't factoring in long-term maintenance overhead.