The question
I run two email products. AgenticBoxes.email is B2A — built for AI agents as customers, with a handful of paying customers to date. Boxes.email is B2C — built for individuals, and it’s brought in about $3,000 in revenue over the last few years against roughly $480 a month in operating costs. That math doesn’t work yet, and I know it.
The honest version of this experiment isn’t “can I build something that makes money.” It’s “can I close the gap between cost and revenue — and is the B2A side the path that gets us there, the B2C side, or both?” I’m betting on the agentic side, but I’m watching both.
Evidence log
May 24 — Agentic onboarding verified in production
A new B2A customer onboarded entirely via API. POST /signup/agentic, POST /signup/agentic/confirm, key returned, authenticated GET /boxes returns 200. No human filled out a form. The whole signup happened agentically, in production, with proper authentication and account setup.
That’s a separate proof point from “the product works.” The product working is verified by customers actively using it. The signup flow working agentically is verified by accounts that came into existence without a human in the loop. Both architectural claims are now true in production, not just in beta.
May 23 — Real-time IoT/MQTT delivery shipped
Every other agentic-email API requires either polling on a timer or a public webhook endpoint. AgenticBoxes now pushes email events to agents in real time over MQTT — agent provisions a client cert, subscribes to its own topic, events arrive instantly. Works behind NAT, on a laptop, inside a sandbox.
The competitive move: from “we have webhooks too” to “we work where webhooks can’t” in one day. The customer pain this addresses is real — sandboxed agents in CoWork-style restricted environments, Claude Code’s scheduled tasks, enterprise agent platforms with outbound-only network policy. These are AgenticBoxes’ fastest-growing customer segments. Real-time delivery without a public endpoint is the architectural feature that unlocks them.
May 23 — Agentic opt-in beta system shipped
Beta features are now exposed to agents the moment they become available. The customer’s agent doesn’t get an email asking them to opt in. The agent receives a broadcast event, opts in via API call, and starts using the beta capability — all without a human reading marketing copy.
One paying customer used this exact path to test a new MCP server within 12 hours of release. Verified opt-in via API; verified usage via real traffic; verified feedback by direct conversation afterward. The whole loop is closed agentically with humans deciding only the on/off, not the click-here.
May 22 — First external paying customer rated production-grade
The first external paying customer tested a new MCP feature overnight, rated it 4 of 5 stars, and identified the one missing piece. His verdict: “Native MCP server works flawlessly in INTERACTIVE sessions. Server, auth, billing, and tool schemas are all correct.”
The customer was a senior technical professional with no incentive to be polite about a half-baked product. The 4-out-of-5 with a specific deficit-call is more credible than 5-out-of-5 enthusiasm would have been. The single deducted star wasn’t a product bug; it was an authentication-scope limit on a feature outside our control.
Across May — Pricing model has structure
AgenticBoxes prices per-message at $0.0004. 250 messages free per account. Top-up is the standard pattern — a few dollars at a time, no minimum. Founders’ Pricing tier (75% off) for early customers.
The cost-per-customer is tracked tightly enough that we know what we’re spending on AWS, Anthropic, and the rest of the stack at any usage level. The pricing equation works at small volume; the question is whether it works at scale where AWS and Anthropic costs scale alongside.
The honest math on the B2C side
Boxes.email has been running for years. Approximately $3,000 total revenue. Approximately $480 monthly operating cost. That’s a real burn rate on a real product with real users — and it predates the agentic work by years.
The B2C side isn’t being abandoned. It’s being held while the B2A side grows into the path that closes the gap. That bet may or may not pay off. I’ll know more in 90 days than I do today.