Builder

AI Builder

Gary designs, builds and ships AI products, taking them from a first conversation about a problem to software that customers run in production.

Two decades of leading enterprise technology taught him what large organizations need before they’ll trust a new system: security, governance, and something that solves a real problem. He now brings that to building products himself, for Caprus AI, where he is co-founder and CTO, and through his own firm, CRIS Advisory.

Products

Brought to Market

Caprus AI

AgentCube

One governed connection between an organization’s AI assistants and its enterprise data.

The problem

Business data sits in warehouses, databases and financial systems that AI assistants can’t reach, and wiring each assistant to each system leaves no central control, no consistent audit, and answers that sound right but aren’t.

What it does

Works with the assistants people already use, including Claude, Microsoft Copilot, ChatGPT and Gemini
Connects to Oracle EPM (Planning, Financial Consolidation & Close and Essbase), relational databases, data warehouses such as Snowflake, and any MCP-compliant source
Turns everyday business language into exact queries, and asks a clarifying question rather than guessing
Each source system’s own security applies to the named user; no business data is stored
Self-hosted in the customer’s own cloud, never a service in the middle

caprus.ai/agentcube →

Caprus AI

AgentCube GO

The application that runs an organization’s AI adoption program, not just documents it.

The problem

Most AI adoption programs live in spreadsheets that depend on one person, and most AI risk shows up after launch, long after the spreadsheet has stopped being updated.

What it does

Staff submit ideas by simply talking to their AI assistant; GO asks only for what’s missing
Every idea is scored and routed by fixed rules, with risk captured at intake and a gate before anything is built
Delivered work is checked monthly, then graduated or retired with a recorded reason
Measures value with a formula that counts the human time spent reviewing AI output
Drafts the executive sponsor’s monthly report automatically

CRIS Advisory

PromptMason

A private prompt library for people who use AI every day.

The problem

Good prompts get rewritten from scratch, scattered across documents, and pasted into public AI tools along with information that should have stayed private.

What it does

Build prompts from reusable blocks; improve a block once and every prompt that uses it follows
Built-in MCP connects the library to Claude, Cursor and other AI apps
A free AI model that runs on the user’s own computer, so private work never leaves it
Encrypted team libraries that sync through the cloud drives teams already use, with no PromptMason server
Desktop app for Windows and Mac, with a companion Chrome extension

promptmason.ai →

Approach

How the Work Gets Done

Start with the problem

Every product begins as a written specification of the problem, the people who have it, and what the product will deliberately not do.

Modern tools, proven discipline

AI-assisted development speeds up the work. Sound architecture, code review and the engineering practices of enterprise IT keep it production-grade.

Ship what holds up

Security, audit and testing are built in from the first commit. The AgentCube products alone are backed by more than 12,000 automated tests.

Have an Idea Worth Building?

Each of these products started as a problem someone could describe in a few sentences. If you have one, a product idea, a process that should be software, or an AI pilot that needs to become something people rely on, Gary would like to hear about it.

Tell him what you have in mind and where it stands today. He’ll reply personally, and the first conversation is about whether and how it can be built, not a sales pitch.