4Why Now
Conversational AI, digitized contractor ops, agentic payments, costly skilled time, consumer expectation. The mandate replaces the checkout: the credential enforces the money limit, orchestration enforces the scope limit. Sensors come later.
~3 min read
| Force | What it changes now |
|---|---|
| AI understands the request | Intake, photos, triage, communication, and documentation become automated. |
| Contractor operations are digital | Calendars, estimates, job histories, and payments are software-accessible. |
| Software can act | Scheduling, approvals, reminders, and payments run with less human coordination. |
| Skilled time is expensive | Less admin means more productive work. |
| Consumer expectations changed | Fast answers, clear status, digital payment, fewer handoffs. |
Sensors are a later accelerator, not the base case: today's installed base is speakers, doorbells, and thermostats, not diagnostic sensors on plumbing and electrical. Near-term entry is conversational and photo-based; telemetry adds signals later, where insurers or manufacturers subsidize it.
Agentic commerce: from checkout to mandate
The payment rails are moving ahead of the category. Visa launched Intelligent Commerce in April 2025 (tokenized credentials, spend controls, authentication for AI-initiated purchases) and followed in October with the Trusted Agent Protocol, so a merchant can cryptographically tell a consumer’s authorized agent from a bot; Mastercard’s Agent Pay runs the same play. Stripe and OpenAI published the Agentic Commerce Protocol in September 2025. In August 2026, Stripe agreed to acquire OpenRouter, a routing layer across 400+ AI models. Visa expects millions of consumers to buy through agents by the 2026 holiday season. Whatever the timing, the direction is clear: identity, permission, and settlement for agents are being built as neutral infrastructure beneath every assistant, not inside any one of them. [14][15]
For homes, this changes the unit of instruction. Today a homeowner buys a job. Under a mandate, they delegate a budget:
“Handle my house this month. Spend up to $1,000. Don’t approve any single repair over $300 without asking me.”
The agent then runs the loop this paper describes (diagnose → find provider → price → schedule → pay) against a credential that is itself programmable:
- Agent
- any authorized coordinator: a vertical service agent, ChatGPT, Claude, Gemini, or whatever comes next
- Scope
- home services
- Monthly budget
- $2,000
- Per transaction
- under $500; above that, ask
- Geography
- local
The network does not need to know which assistant did the work; it enforces the money boundary. What it cannot enforce is the scope boundary: the difference between a $180 drain clearing and the sewer line it reveals, or the AC extension whose main unit had no capacity (Section 5). That judgment lives in the orchestration layer. Programmable credentials make the front of the transaction trivial to delegate, which makes the layer that decides what is worth paying for the one that matters. It is the thesis, restated in payments.
Software stops helping people coordinate the job. It starts coordinating the job.