Consulting
From the first question to agentic operations.
I help leaders in government, public health and enterprise decide where AI belongs in their operations and plan how to get it there. You leave with a strategy your budget can support and your security officers can approve, one your own teams can carry out. It runs from the first secure sandbox to AI agents at work across your organization.
Starting points
Start where you are.
Finding where to start
You know AI matters to your work, and you want to know where it fits first. I help you set a clear goal and pick the first use cases worth funding. Then I show your team how to set up a secure sandbox in your own environment where staff can try approved tools without exposing your data.
Planning for growth
Some of your teams already use AI, and leadership wants more from it. I help you decide what to scale and forecast what that growth will cost. Before adoption spreads to more teams, I review your policies and governance and tell you what to change first.
Moving to agentic operations
You are ready for AI agents that carry out work on their own. I advise on the architecture and governance that keep those agents secure on infrastructure you control, including where a person on your staff should approve each decision with real consequences.
The process
Five steps to a realistic plan.
I follow the same five steps for a city department and a large enterprise. The size of your organization and how far along it is decide how deep each step goes.
Set the North Star
I start with your leadership team and the outcome they want from AI, then help them turn it into a goal specific enough to guide decisions and to measure progress against. Every recommendation I make from then on has to serve that goal.
Measure true readiness
I look past the plans on paper and ask the questions that reveal how ready your organization is in practice. Those questions cover four areas.
Policy
Your acceptable-use, data and procurement policies, and the changes they need before they cover AI.
Governance
The people who approve AI use, and the way your teams assess risk and review decisions.
Staff
Your people’s skills and comfort with AI today, and the training they need to use it with confidence.
Infrastructure
Your identity, data, network and cloud environment, and the work it needs before it can host AI under your security controls.
Find the use cases that pay off
Many processes could use AI, and only some will repay the investment. With your teams, I weigh each candidate’s value against its risk and effort. For every use case you choose, I recommend the baseline your team should record before launch so you can measure the gain.
Budget and forecast
AI costs grow with adoption. Licenses, model usage, cloud capacity and the staff time to run it all rise as more people pick up the tools. I sit down with your finance team to check the budget and forecast against realistic usage and to plan for the growth that follows a successful pilot.
Climb from sandbox to agentic operations
The strategy takes the form of a safe ladder your organization climbs one rung at a time. It starts with a secure sandbox where staff try approved tools and rises through pilots in production to an organization that runs AI and agents in its everyday operations. Each rung has its own security controls, budget and measures of success. You climb to the next rung only after the one below has met its targets.
Security and risk
Security that keeps pace with adoption.
I make risk protection part of every step in the plan. For each use case, I set out the safeguards it needs and how your teams can put them in place before anyone on your staff relies on it.
Threat modeling
Before an AI system goes live, I walk your team through how someone could misuse it or how it could fail: the data it can reach, the actions it can take and the people who might try to manipulate it. Your team leaves with a written list of those risks, each with an owner in your organization and a control to match.
Data protection
I advise on where your data should live, which people and agents should read it and what may leave your environment. Then I explain to your security team how to enforce those decisions with encryption and access controls, and how to test each one.
Input protection
An AI system can mistake the text it reads for instructions. In a prompt injection attack, someone hides commands in an email or a web page for an agent to find. I teach your engineers to place input filtering, content isolation and narrow tool permissions between your agents and anything untrusted.
Model and vendor risk
I assess where each model and vendor you depend on runs, how it handles your data and what happens if it changes its terms or goes down. For each critical service, your plan names a fallback and the steps to switch to it.
Oversight and audit
I work with your team to define the limits each agent works within and the actions that need a person’s approval. I also recommend the logging your auditors need to trace any agent action back to the request and the data behind it.
Without slowing down
A manual security review for every change creates a backlog. I point out which checks your teams can automate with policy rules and security scans and where those checks belong in the pipelines they already run. Changes that pass them keep moving, and your security team saves its time for the changes that need a person’s judgment.
Your teams
AI for the people doing the work.
I advise three groups in your organization on the AI tools their jobs call for and how to bring them into daily work.
Office and administrative work
Most of your staff will meet AI first in email, documents, meetings and research. I help you choose assistants such as Microsoft Copilot and ChatGPT for that work, and I review your IT team’s configuration against the data rules in your plan.
Data science and analysis
Analysts and data scientists gain the most when AI works inside the tools they know. I suggest where it fits in their notebooks, models and dashboards, and I coach them to build each workflow with tests and versioned data so an auditor can rerun it and get the same numbers.
Software development
I brief your engineering leads on current AI coding tools such as Claude Code and Codex, and on the checks code written by agents should pass before it ships. Those checks belong in rigorous CI/CD pipelines that test and review every change before it merges, and I teach your developers how to set them up.
Background
Hands-on since the early days.
I have worked with generative AI since the early days of ChatGPT and watched it grow from chat assistants into agents that write software and run operations. I specialize in municipal government and public health, where I design agentic systems, run AI operations and train public-sector teams to use AI in their work. That work handles residents’ data and answers to the public. I bring the same standard of security and accountability to every engagement.
I write my own software as well as advise on it. I have built SaaS applications across full technology stacks: machine learning and dashboards in Python, web apps in React, TypeScript and Node, and agent tooling in Rust, C and C++.
Most of my public-sector work runs on Microsoft’s cloud and AI platforms, and I use the agentic tools from OpenAI and Anthropic in equal depth. I also use open-source models and tools. When your systems live in Amazon Web Services or Google Cloud, I adapt the plan to those services. My advice on what agents can and cannot do comes from running a fleet of them in my own studio every day.
Platforms
- Microsoft Azure
- Microsoft Foundry
- Microsoft Copilot
- ChatGPT
- Claude Code
- Codex
- Open-source models
- Amazon Web Services
- Google Cloud
Tell me what you want AI to do for your organization and where you stand today. I read every message myself, and I’ll reply with what your first step could look like.