Base Batches 003 · Business Summary
ChatGPT answers. Mia gets it done.
Relationship-driven consumer AI operator that turns trust into execution.
Three events in a single month made the shift obvious.
People don't lack information. They lack completion.
Everyday life is full of small but costly unfinished decisions: which app to use, whether to reorder, whether to spend, whether to wait. These decisions are easy to delay, fragmented across apps, and expensive to get wrong.
AI assistants can answer almost anything. But trusted completion is still missing. The hard part is not intelligence alone — it's permission, confidence, and context.
You talk to Mia. She makes the decision with you, then executes through existing apps and rails — not inside a proprietary checkout.
ChatGPT Checkout tried to make you buy inside the chat. That's the wrong abstraction — it fights how people actually shop. Mia doesn't replace your apps. She sits above them — deciding which one to use, when, and at what price, then routing execution through existing rails.
Over time, general-purpose assistants will also execute tasks. But execution is not the moat. Trusted delegation is. Mia is optimized around your boundaries: what you approve, what you delay, what you regret, and what you want automated.
OpenAI builds the smartest assistant. We build the most trusted operator. That requires a relationship, a policy layer, and a history of successful completion.
High frequency · fragmented apps · repeated decisions · real money at stake · fast feedback loop
| # | Source | Description |
|---|---|---|
| ① | Subscription | Pro $15/mo — deep memory, priority execution |
| ② | Execution fee | Fee on completed tasks. Works on free users too. |
| ③ | Merchant fee | Platform take rate when Mia completes a merchant-side checkout flow |
Phase 2: Paid tools margin (MPP/x402) · Phase 3: Financial margin (balance, card, payouts)
Users never see the rails. They just see Mia.
Conversation, memory, preferences, budget sense, trust. This is the product.
MCP connectors, merchant APIs, aggregators (e.g. MealMe), bookings, comparisons, completion workflows.
Merchant checkout through agentic commerce rails.
Tool payments through machine-native payment rails.
Onchain execution and onboarding where useful.
Base is the enabling layer, not the product. The product is Mia.
Base gives us the fastest stack to test this: onboarding, agent-native execution, and programmable payments in one ecosystem.
The real chokepoint in the agent economy is not just payment. It's permission, policy, and trusted completion.
We don't wait 100 actions to learn you. With permission, we import existing digital behavior — calendar patterns reveal time delegation, payment history reveals spending boundaries, response speed reveals relationship priority.
Day 1: import → infer delegation hypotheses → confirm with user → ready.
We don't present imported data as truth. We turn it into delegation hypotheses, then confirm them with the user. The trust comes not from reading more data, but from making the confirmation loop explicit.
In March 2026, the market made one thing clear: agentic commerce needs more than intelligence. It needs trust, permission, and completion.
The lesson: payment infrastructure is getting real. Trust infrastructure is still missing.
Northworld is a relationship-driven consumer AI operator. Mia builds trust through repeated, everyday delegated decisions, then uses that trust to complete tasks with money at stake.
The moat is the delegation OS — a personal permission rail that knows what each user will approve, reject, or automate. No general-purpose AI can build this. It requires a relationship.