I have sold it, built it, governed it and run a business on it. Pick the door that matches your company and start there.
The moment AI is reading, writing, deciding and acting across a business, the fabric exists. Nobody bought it or installed it. And almost everywhere, the only thing standing between what an agent intends and what it does is a policy written in text that no running system has ever read.
I build for where buyers will be in eighteen to twenty-four months, because what they want today, everyone is already selling. In 2024 few buyers were asking for mathematically governed AI. I went after it because I expected the probabilistic nature of these systems to become the question on every serious enterprise deal, and today it is.
I founded SMARTHAUS to make that fabric governable: rules proved before they run, enforcement at every place an AI can act, and a record that survives the argument.
I am not an engineer, and that turned out to matter. I developed a framework, Mathematical Autopsy, that lets me develop code through a math-first, formal-methods process in collaboration with AI. The rules a system must obey are proven before any code exists. AI is free to propose, and the proof decides.
The idea underneath it: correctness cannot come from a probabilistic model, so the rules are defined outside the model, proved, and enforced before the AI acts, with a record of every decision. Three provisional patent applications sit behind the work. Source: USPTO provisional filings 63/942,361 · 64/030,463 · 64/030,568
So I can sit with your engineers and your buyer’s architects and still bring the answer back to the CFO in plain terms.
Prospect Forge is a tool SMARTHAUS built for revenue teams that sell AI. It finds accounts from hiring signals in public job boards, researches each one, writes an opener grounded in what it found, runs every draft through an automatic gate (verified addresses only, required compliance lines, sending limits), and sends nothing without a person’s approval. A company can put it to work today.
I work strategic accounts myself with the same discipline: AI reads the filings, earnings calls, news and job posts, every figure in an account plan carries its source, and anything I cannot verify is labeled a hypothesis. A plan built like that holds up when an executive asks where a number came from.
I was early here too: in 2023 I co-led the AWS initiative behind the integration of MongoDB into Amazon CodeWhisperer, an AI coding assistant, before AI coding had a popular name.
Source: MongoDB and AWS press release, 6 Nov 2023Every enterprise buying AI agents asks two questions: what is this agent permitted to do, and how will we know what it did? A policy wrapped around a model after the fact will not answer them in front of a regulator, a board or a customer.
SMARTHAUS is the IP company I founded to answer them. The Universal Control Plane decides before the AI acts, refuses what it may not do, and writes a receipt either way. You can try a small version here: the Ask about Phil assistant may only quote my approved record.
Further reading: You Cannot Govern a Guess With Another Guess and The Gates Between the Seams.
Every company is asking how to put AI to work for its own people, not just in a chat window. My answer is an Employee Command Center: the next generation of the company intranet, with AI built into it. It sits on top of the tools people already work in and connects, under the company's own rules, to each employee's own AI agent. The people drive the experience, and the company can prove what every agent did. Each kind of team needs something different:
Pipeline that fills itself without spamming anyone, research a seller can trust, and a clean handoff the moment a real person replies.
Agents find accounts from hiring signals, research each one with every fact sourced and draft the first note. Nothing goes out without a person's approval, and a person takes over at the first reply. SMARTHAUS built this as a tool, Prospect Forge, that a revenue team can use today to sell AI.
One view of every owned company on the firm's own value-creation levers, numbers a board can trust, and nothing about one company leaking into another.
Each owned company is its own compartment, every number carries a receipt back to its source, and the board packet leaves only when the CEO sends it.
The other side's production turned into a working record, client data that never leaves the firm, and the firm's AI policy actually enforced.
The work runs on the firm's own machines, the evidence-review workspace sits at the center, and the firm's AI policy is the product's settings.
I have built working examples of the holding-company and law-firm versions on fictional companies, and I walk through them in conversation.
Each part of a modern AI system works. The joins between them do not. My thesis proposes mathematics itself as the integrating substrate, and it is falsifiable by construction: every implemented guarantee maps to a measured invariant.
Read the thesis (PDF, 45 pages)