The Ledger.
A record of AI systems I’ve built — for clients, as products, and in the lab. Names are withheld and details are redacted to protect the people who paid for the work. What’s left is the thinking, and what it would have taken before AI.
A consistent pattern runs through all of it: surface signal from noise, generate artifacts instead of maintaining them, and put outputs where people already work — not in yet another dashboard.
Closed loops, not one-shot agents
The best work isn’t a tool that answers once — it’s a system where every human interaction makes the next answer better.
Generated, not maintained
No one updates a board. Artifacts are produced from the natural signals of real work.
Capture at the point of work
Inputs happen in Slack, Notion, the inbox — wherever the team already lives. No context-switching.
Accuracy as a constraint
Every claim labeled verified, inferred, or estimated. A missing field beats a fabricated one. No claim outruns its data.
Human-in-the-loop by default
Every output is reviewable. Templates leave placeholders instead of inventing facts.
Graceful degradation
Components fail quietly. Publish what you can, skip what you can’t.
Hybrid systems appear in both groups.
A marketing department's worth of engines
A set of standalone AI engines — one writes campaigns, one finds a competitor's customers, one builds and improves its own landing pages — each handing a small marketing team finished work it can use the same day.
→ Marketing work done by hand across several teams → AI engines that hand the team finished work where it already works.
An AI operations layer for a business too small to staff one
Scheduling, marketing, sales follow-up, technician training, ads, and profit analysis — AI woven through the daily running of a small, owner-operated field-service business.
→ Jobs a small business could never afford to staff → an operations layer one person runs.
A real-time call coach that runs on your laptop
A Mac app that sits around the notch and coaches you live during a video call — surfacing your talking points, flagging when you're talking too much, reading the room — without sending anything off your machine.
→ Cloud transcription, lag, and a privacy problem → a private live coach that runs on your own laptop.
An agent that answers every lead in seconds
A new inquiry gets a real, helpful reply by text or email within seconds — any hour — answering questions, sorting out who is worth pursuing, and moving toward a booking, with no one watching the inbox.
→ A staffed desk racing a five-minute window → an agent that always wins it.
A voice agent that books the appointment
A voice agent answers the phone for small service businesses, holds a natural conversation, and books the appointment — one engine quietly serving many businesses at once.
→ A receptionist or answering service per business → one voice engine serving all of them.
A personal operating system that runs on your own machine
Contacts, projects, email, calendar, a meeting analyst, a decision journal — all working together, all stored on your own computer, all running on the Claude subscription you already pay for. No servers, no per-seat fees.
→ A stack of monthly SaaS subscriptions holding your data → one local system you own.
Sites that generate themselves from public data
The pipeline finds a business, builds it a real website from information that's already public, and puts a live preview link online — no designer, no developer, no kickoff call.
→ An agency building each site by hand → a pipeline that generates and ships them.
From demand signal to shipped product, on a pipeline
One pipeline runs the whole idea-to-launch funnel: it scans the open web for signs of real demand, weighs which ideas are worth it, builds the promising ones, and works out how to take them to market.
→ A team working the idea-to-launch funnel over months → one pipeline that runs the whole funnel.
A sales intelligence system that learns from every call
Paste a potential customer's website and, two minutes later, the salesperson has a short, fact-checked prep sheet — who they are, whether they're worth pursuing, and how to approach them. Paste the call transcript afterward, and the system gets smarter for the next one.
→ Every call prepped from scratch → a system where the team's hard-won sales knowledge builds up on its own.
Voice-of-customer intelligence, on autopilot
It reads what customers say in their own words — across forums, review sites, online groups, and video — and watches what competitors are quietly doing, then writes it up. Every week, on its own, instead of once a year by hand.
→ A research team's once-a-year sprint → a continuous loop one person runs every week.
Getting cited by the AI that answers for you
When someone asks ChatGPT, Perplexity, or Google's AI for the best tool in your category, does it name you? This measures that — and shows exactly where you're invisible and how to fix it.
→ A question nobody could answer two years ago → a repeatable health score and a fix-it roadmap.
A marketing department's worth of engines
A set of standalone AI engines — one writes campaigns, one finds a competitor's customers, one builds and improves its own landing pages — each handing a small marketing team finished work it can use the same day.
→ Marketing work done by hand across several teams → AI engines that hand the team finished work where it already works.
Business reviews that compose themselves
Each leader answers a 15-minute AI interview, and the monthly business review builds itself into a finished, on-brand deck — current, not a stale snapshot, with nobody hand-assembling slides.
→ A multi-day slide-stitching marathon → reviews that build themselves from real answers.
Team visibility nobody has to maintain
The team drops a short, tagged line in their chat tool when something happens — shipped, learned, dropped — and the weekly leadership update, the project cards, and the metrics build themselves.
→ Status meetings and boards someone has to keep current → a leadership update that writes itself from the team's normal work.
An AI operations layer for a business too small to staff one
Scheduling, marketing, sales follow-up, technician training, ads, and profit analysis — AI woven through the daily running of a small, owner-operated field-service business.
→ Jobs a small business could never afford to staff → an operations layer one person runs.
A personal operating system that runs on your own machine
Contacts, projects, email, calendar, a meeting analyst, a decision journal — all working together, all stored on your own computer, all running on the Claude subscription you already pay for. No servers, no per-seat fees.
→ A stack of monthly SaaS subscriptions holding your data → one local system you own.
The harness everything else is built with
A reusable kit of building blocks — proven patterns, recipes, quality checks, and accuracy guardrails — so every new system starts where the last one ended instead of from scratch.
→ Re-building the same plumbing on every project → every build starting from an accumulated system.
An autonomous company, orchestrated by agents
Set a goal, and a 'company' of AI agents — each with a job — divides the work, hands it off between them, and produces the result with little steering from a person.
→ A research demo of agents coordinating → a system you can actually run.
Five specialized agents that remember
Five AI agents, each with its own job and a memory that lasts — so they pick up where they left off and build on past work instead of starting cold every time.
→ Stateless one-shot agents that forget → a standing team of agents with memory.
The redacted parts are the parts clients pay for. If you want to see a live system and talk through what it would take for yours, that’s a conversation.