AaaS · internal briefing · for Tobias
An email agent that sells email agents.
Over email. On autopilot.
This is the whole company in one sentence. Our product is email agents. Our growth engine — the GTM Fleet — is email agents whose job is to sell email agents. The ad and the product are the same thing. When it runs, it markets itself, closes itself, and pays for the next agent it builds.
01 — the product
What an "email agent" actually is
Raw GPT / Claude / Gemini give you a smart text box. To make them do anything, you wire up APIs, tools, MCP connections, auth, data — yourself. That's the work. That's the friction.
AaaS removes it. You pick a micro-agent — a prompt that only becomes powerful with the right API / tool / MCP wired behind it — and our system is already hyper-connected. No setup, no keys, no plumbing. You email a request; the agent executes on autopilot; you're charged per request. The value isn't the prompt — it's the connection layer we run so you never touch it.
Everyone else
A chat model + your weekend
- ×You wire the APIs, tools, MCP, auth
- ×You set up and maintain the data sources
- ×You babysit the run, glue the steps
- ×Pay a flat seat whether you use it or not
AaaS email agents
Send an email. Get the outcome.
- →Zero setup — the connections already exist
- →AaaS provisions every relevant data source
- →Runs on autopilot — the agent does the plumbing
- →Pay per request — usage is the meter
02 — the insight that changes everything
The product is the ad
Here's the question that unlocks the company: what is the most convincing way to sell an email agent? Not a landing page. Not a demo call. Have it email you — and do the thing.
Every cold email we send is a live email agent. The prospect doesn't get an ad for the product. They get the product.
They reply — and the reply is the activation. The same inbound runtime that serves paying customers picks it up, runs the agent, delivers the result. No handoff, no "book a demo." The moment of interest and the moment of first value are the same keystroke. That's a conversion mechanic nobody with a normal funnel can copy — because for us, the funnel and the factory are one machine.
03 — the self-promoting loop · seven micro-agents
Every stage is an agent. Every agent already renders.
To promote itself, the system needs a fleet of micro-agents that run the whole go-to-market motion end to end. Below is that motion — and the live dashboard for each stage. The shop window is built. These are real screenshots from canvas.aaas.love, today.
Find the next agent to sell
Researches which micro-agent the market wants next, and scores it. The fleet decides its own roadmap.
Define exactly who buys it
Firmographics, personas, triggers — and a compliance-critical country gate. Targeting becomes machine-queryable.
Find & verify the contacts
LinkedIn + email discovery, then deliverable / risky / catch-all / undeliverable buckets. A clean, compliant list.
Write the offer, per lead
The email that promotes the agent — personalised to each prospect's pain. Message → pain, traceable.
Send it — as the product
Warmup, throughput, deliverability. Each send is a working agent, one reply from activating.
The reply activates the agent
The prospect replies; the inbound runtime runs the offered agent and qualifies them — no demo, no wait.
They convert — and become the next promoter
Usage-based billing kicks in. Every new customer is another live demo of an agent in the wild — which feeds the top of the loop again. Revenue funds the next agent the Scout finds.
04 — why it compounds
It's not a funnel. It's a flywheel that spins itself faster.
A funnel is something you keep pouring money into the top of. This refills its own top. Four couplings turn a linear pipeline into a compounding engine:
Improves itself
Decision-traced A/B promotion. Winners get ratcheted in automatically.
Feeds its own content
Schedules the surrounding content + review-gate, so the outbound never runs dry.
05 — the honest status
The shop window is up. The engine room is yours to build.
This is the part that should excite you, not worry you: the hard, ambiguous product-design work is done and live. Eleven fleet dashboards render on canvas.aaas.love right now. What's missing is the thing you're best at — the backend that makes the buttons do the work.
The surfaces + the inbound runtime
All 11 GTM dashboards render (you've seen them above). And the core product loop already works: email → agent-router → gemini → reply + canvas is wired in functions/src. The thing we sell exists.
The GrowthRun engine + the bridge
No state-machine sequences the stages yet; the real runners (lead-scan, personalize, Stripe) are real but unwired islands; and the cold-reply → activation bridge doesn't exist. That's the backend. That's you.
06 — the ask · what you build, Tobias
You're not building a marketing tool. You're building the machine that builds the business.
And the best part: it runs on the exact product we sell. The backend of the growth engine IS an application of the email-agent runtime. Build it once, and you've hardened the product and the go-to-market at the same time. Concretely, in order:
The acquisition runners — you're ~60% there already
Extend the real icpLeadScannerRunner.ts: a compliance-grade ICP + country-gate, bulk LinkedIn (the Apify actors are already defined), and SMTP-ping verification. Output: clean, verified, compliant target lists. No sending. → issues #3632 / #3633 / #3634
The GrowthRun spine — the state machine
Define defining → building_audience → outbound → converting → done and wire the existing real runners into it, so the fleet stops being islands and starts being a pipeline that hands off artifacts. Reconcile the two data stores so the dashboards show real runs, not seed.
Send + compliance (gated)
Cold-send infra on its own domain (never the transactional one), behind a real C0 gate — suppression, country-gate, complaint auto-pause. This is the one place we move slowly and correctly. Nothing sends until it's compliant + signed off.
The reply → activation bridge — the keystone
This is the whole thesis in one function. A prospect replies to a cold email → dedupe it → map it to the offered micro-agent → create an agent_task → the existing inbound runtime executes and delivers. The moment you close this loop, an outbound email becomes a self-activating product, and the funnel becomes the factory. Everything else exists to make this one bridge fire.
None of this is speculative. Every stage has a live dashboard, a data model already typed (growth-run.types.ts), and a design spec. You're not inventing the shape — you're powering a machine that's already standing.
Build the bridge, and the product starts selling itself — literally.