Everyone wants to use AI.
I make it drive real business results.
The gap between what AI can do and what your company actually does with it is the largest arbitrage in business right now — and it doesn't close with strategy decks. It closes with working systems that move one of three numbers.
CTO, Curbily · TRM Labs alum
founded one of the first US Bitcoin funds (2012)
Don't take my word for it — touch it
Most AI consultancies show you slides. These are live systems you can click right now, built with the same engine I install in client companies.
The dependency graph
A chipmaker's full supply chain as an explorable entity graph. Click any company, drill into what it depends on, follow the sole-source risk upstream to the mine it all starts from.
Explore it → ● live money, live nowThe market graph
Pulls the highest-volume open Polymarket contracts the moment you load it and links them into an entity graph in your browser: pick Powell, Ukraine, or Bitcoin and see every market exposed to them — and who they're entangled with. Also an MCP endpoint your AI can query.
See who moves the money → ● real companies, live pricesThe stock dependency graph
The fictional chain, real: Nvidia → TSMC → ASML → Zeiss — three actual sole-source links under the most-owned stock on earth, with live prices on every public node. The risk lives left of the ticker everyone watches.
Walk the real chain → ● anonymized from productionThe creator graph
Which creators should a brand book? The matching layer under a real 2,900-creator marketplace, with invented names: briefs ranked against scored profiles, "don't book both" audience overlaps, and the agency paths to the talent.
Match a brand →The relationship graph
Point it at a Gmail account and it maps who knows who: 306,000 of my own messages became 3,700 scored relationships — who's warm, who went cold, who can introduce you to anyone.
Get it for your inbox →Case studies
Every engagement is named for the number it moves. These are mine — run on my own companies first, the same way I'd run yours.
The operating graph: what a restaurant CEO's data should look like
Darden Restaurants — 11 brands, 2,159 restaurants, $12.1B — as one clickable graph built from its 10-K: which brand earns its capital, where the next 60 openings go, what moves COGS (with live cattle futures on the beef node), and the decisions hiding in plain sight. Public data only; the installed version runs on yours.
The dependency graph Red Lobster didn't have
America's biggest seafood chain went bankrupt with 100,000 creditors — and the autopsy reads like a graph query: an owner who was also the sole shrimp supplier and a creditor, plus a landlord created by its own buyout. All of it public, none of it joined. I built the graph from the filings.
Found 95% of AI spend hiding in one workload
Curbily's model spend looked like a flat monthly bill. Attributing every dollar to specific workloads showed 95% of it concentrated in one unattributed job — and the true cost-per-user-action made the existing pricing untenable.
Idea to strangers paying in 72 hours
A live multiplayer iMessage battle game: AI judge, image renders on players' real selfies, Apple Pay purchases in-chat. Built, shipped, and taking money from strangers within 72 hours of the idea.
2,900 raw roster rows became a scored creator graph
Agency rosters arrived as messy lists. The Entity Engine turned them into canonical creator profiles — photos, stats, scores, media kits — deduped, enriched, and refreshed nightly for $40.
Services
Fixed price, fixed scope, delivered as working software — because "AI transformation" you can't see on a P&L didn't happen. I run my own company this way: Curbily operates with 15 AI agent roles built and run by me. Barnett Labs installs that capability in yours, two engagements at a time.
① The AI Cost & Margin Audit $15,000 · 5 business days
For AI products already live. I attribute every dollar of your model spend to specific workloads, compute your true cost-per-user-action, stress-test your pricing, and hand you the caps and fixes.
② The MVP Sprint $35,000 · 1 week + 1 week of fixes
Idea → working product with real users on it. Not a prototype — deployed, instrumented, taking payments if it should. See the Slop Fight case study above.
③ The Entity Engine from $50,000 · 2–3 weeks
Your messy lists, rosters, or CRM become a canonical, scored, continuously-refreshed entity graph — with dedupe and resolution, connectors, and an agent-ready MCP endpoint on top. Graphs are plumbing; scores are decisions.
④ The Agentic Inbox from $60,000 · 3–4 weeks
An AI agent that lives where your customers already are — email or iMessage — triaging, answering, routing, and closing. Includes abuse guarding, human handoff, and analytics. Six production inbox agents in daily operation, plus the only iMessage-native game bot in existence.
⑤ The AI-Native Operating System from $120,000 · one quarter
The full install: your company restructured around agent workflows — engineering velocity, QA harnesses, content pipelines, ops dashboards, cost instrumentation — plus your team trained to run it. This is the Curbily company-OS, productized.
⑥ Fractional AI CTO $20,000/mo · 3-mo minimum
Everything above, continuously, with my name on your problems. Two slots exist; one is open.
Why these prices are cheap
A single senior AI engineer costs $350K+/yr fully loaded, takes three months to hire, and arrives without a system. Agencies quote MVPs at $50–100K and deliver in a quarter. The audit routinely finds five figures of annual burn in week one.
The lab takes two builds a quarter — Curbily is my day job, and that scarcity is your guarantee that I'm doing this for the craft and the compounding, not volume.
Start with 30 minutes
A diagnosis call, not a sales call: we look at where your AI spend or AI roadmap actually stands, and you leave with at least one thing to fix — whether or not we work together.
Book the AI P&L review