I help enterprise teams design, build, and govern multi-agent systems, from strategy through production. No vendor allegiance, no slide-deck consulting. The deliverable is always something your team can run after I leave.
Looking for advisory only? See /advisory. Looking for training? See /education. This page is the full menu.
Models, APIs, MCP servers, vector stores. The raw capabilities. Most companies overspend here and underdeliver because the layer above is missing.
Encoded workflows. Reusable prompts, tested chains, named playbooks. The middle layer where one engineer's experiment becomes the whole org's capability.
Persistent, role-defined systems that operate in the business. Audited, governed, observable. The output most companies want; the layer most can't yet build alone.
Every engagement starts by mapping where you are on this stack. Then we work the gap.
Audit of your existing agent stack: orchestration model, tool boundaries, eval setup, observability, failure modes. Deliverable: written report with prioritized fixes and a reference architecture for v2.
fit: Teams shipping agents in production who suspect they're hitting a ceiling.
Stand up the policy, process, and tooling for governing agents at scale. PII filtering, alignment checks, audit logging, risk tiering, model approval workflows. Tailored to your regulatory posture.
fit: Regulated industries, enterprises with legal-flagged AI use, or teams under board scrutiny on AI risk.
Live, custom briefing for your C-suite or board on the current state of agentic AI: what's real, what's marketing, and where the opportunity is for your industry. No vendor slides.
fit: Leadership teams that need to make a multi-million-dollar agent bet and want a non-vendor view.
Hands-on training for your engineering, product, and operations teams on building with Claude, Codex, and Gemini. Working from your real problems, not toy examples. Includes a take-home toolkit.
fit: Teams that have read the docs but haven't shipped, or that need to scale agent expertise past one or two early adopters.
I join your team for a focused build. Pair with engineers, set architecture, write the spec, ship the prototype. The goal is a production-grade agent + clear playbook your team can extend.
fit: Teams with budget and urgency but a capability gap: need to ship a flagship agent in a quarter, not a year.
Speaking engagement for your conference, all-hands, or client event. Pick a topic from /speaking or commission a custom angle.
fit: Conference organizers, internal events, customer summits. Past audiences: AI4, POSSIBLE, MMA, TEDxNYU.
Three modes. Each maps to different problems and team realities. Toggle to see which fits.
A custom briefing for your C-suite or board on the current state of agentic AI: what's real, what's vendor pitch, and where the opportunity is. No slides from a deck library.
A monthly seat at your decision-making table. Async input on architecture choices, RFPs, vendor reviews. Monthly working sessions. Slack channel I actually read.
A senior outside read on your existing agent stack. Orchestration model, eval, observability, failure modes, governance posture. Written deliverable with prioritized fixes + reference v2.
I join your team for a focused build. Pair with engineers. Set architecture. Write the spec. Ship the prototype. Leave you with a playbook your team can extend after I go.
Stand up the policy, process, and tooling for governing agents at scale. PII filtering, alignment checks, audit logging, risk tiering, model approval workflows. Tailored to your regulatory posture.
Conference keynote on AI agents, responsible AI, regulation, or bias/fairness. Custom-tuned to audience. Recording rights negotiable. Travel as needed.
Hands-on training for your engineering, product, ops teams on building with Claude / Codex / Gemini. Working from your real problems, not toy examples. Take-home toolkit.
Rates use a 4-tier scale ($–$$$$). Exact numbers depend on scope, urgency, IP, and travel. Email for specifics.
If you're betting the next budget cycle on agents and need a hands-on operator, not a deck, let's talk.
$ schedule a call