The One-Person Company — OPC Reference Architecture

The OPC — One-Person Company — is di-factory's reference architecture for turning a validated Business Plan and a working MVP into a running company operated by one human founder and a team of AI agents deployed in the founder's own cloud. It is the activation layer, the bridge between 'I built something' and 'I'm running a business.' The hero visual puts the founder-CEO at the exact center of the diagram, wraps them in a 3-phase delivery arc — Adjust (2 weeks · $2K) → Launch (4 weeks · $4K) → Accompany (4-week sprints · $3K avg) — and floats every layer of the stack as a swappable chip on the outer ring. The founder owns the center; every vendor around them is visibly detachable. That is the pitch and the architecture, encoded in one image.

A blue return arc from Accompany back to Launch/Adjust makes the iteration loop literal — no lock-in, no annual contract, monthly renewal, stop anytime, no penalties.

The agent roster is dynamic. Every OPC gets a C-suite sized to its business — no fixed template. The diagram shows a reference default (CGO · COO · CTO) as three solid agent nodes — the minimum viable C-suite for a founder-led company. The dashed expansion strip below (CFO · CMO · CDO · CSO · CPO · CHRO) shows the roster grows on demand as the business grows: add a CFO when finance gets complex, a CMO when marketing needs a dedicated lane, a CDO when data becomes a discipline. And every C-suite agent runs its own team of specialist sub-agents underneath — a CGO would have a content agent, a CRM-research agent, a deck renderer, a social drafter; a CFO would have a reconciliation agent, a forecasting agent, an audit-prep agent; and so on. One founder, a variable-size C-suite, layered sub-agents beneath — that is the OPC operating model.

The stack is deliberately swappable at every layer. Runtime: OpenClaw is the recommended agent OS, with Hermes (Nous Research, MIT) and Paperclip (Paperclip AI, MIT) available as peer open-source runtimes on request. Coding agent: Claude Code recommended, Codex and Cursor supported. LLM: Anthropic recommended, plus OpenAI, Google, xAI, and local (Ollama / vLLM) for data-residency or air-gap cases. Cloud: AWS-first (Terraform module ships to AWS in v1), GCP and Azure on the roadmap. Human-agent collaboration surface: Multica (MIT, self-hostable) recommended as the founder's inbox onto agent output — issues, workspaces, comments — with the OpenClaw dashboard (GA roadmap), Linear, and GitHub Issues as alternatives. Below the diagram, four tabs render the full stack against OSS, AWS, GCP, and Azure — so procurement, security, and finance see their preferred vendor path without asking us to redesign.

Commercial Principle #1 is baked into every row of the stack: di-factory charges labor only. The client pays every third-party vendor directly — cloud, LLM APIs, SaaS licenses, all of it. Zero markup, zero rebate, zero commission. The client always chooses the vendor. Every OPC contract includes a Cost Transparency Clause where the client sees third-party invoices raw. Trust wedge, made structural. The four use cases below — solo consultant, freelance founder, digital nomad, boutique advisory — show the OPC pattern in the segments we sell to hardest across LATAM.