Work with me

From a one-hour call to a finished product.

AI agents, web apps and SaaS, voice agents, generative media and 3D, designed and shipped end to end. Or just advice, if that is all you need. You work with the engineer who writes the code, not an account manager.

Start a buildSee what I have built

Ways to work together · start anywhere

Advice · 1 to 2 hours

Consultation

Pressure-test an idea, choose a stack, or review an app or AI plan. You leave with written next steps.

Fixed fee per session
Step 1 · 1 to 2 weeks

Discovery sprint

We map the job, I prove the hardest part works, and you get an architecture, a fixed quote and a working spike.

Fixed price
Step 2 · 3 to 8 weeks

Build

The system, in production, on your own cloud account. Guardrails, monitoring, a runbook and the source code are yours.

Fixed quote from discovery
Step 3 · Monthly

Run and improve

I watch it, tune it and extend it a few hours a week, so it keeps working after launch.

Monthly, agreed up front

Look inside what is running

Pick a system to see how it flows, what it runs on, and the rules that keep it safe.

Pantheon · an agent crew that runs a portfolio

Running

A chief of staff and five specialist agents (product, marketing, infrastructure, engineering and research) work one shared board for every venture I run, and report back without waiting on me.

How it flows

  1. 01BoardPlans and tasks for every venture, filed by people or agents.
  2. 02DispatcherStarts each run as its own process, so one crash never takes the rest down.
  3. 03Agent lanesEach agent works in its own lane with its own charter and tools.
  4. 04WatchdogEvery 5 minutes it checks the dispatcher and restarts it if stalled.
  5. 05Nightly canaryOne real job runs end to end every night and after every deploy.
  6. 06ReportOutcomes are filed back to the board as a record.

Runs on

  • Claude Code
  • TypeScript
  • Next.js
  • pm2
  • AWS

Guardrails built in

  • Memory and time caps on every run
  • Usage pause when the model allowance runs low
  • Cloud permissions deny spend, deletes and permission changes
  • Decisions that need a human go to a card, not a guess
Read the full case study

What people bring me

A web app or SaaS product

An MVP, an internal tool, a customer portal or a rebuild of something that has outgrown itself. AI only where it earns its place.

One workflow, automated

A report, an inbox, a data entry job. Small, fixed scope, quick to pay back.

A team of agents for one job

Research, reporting, triage or ops work that runs on a schedule and reports back.

A voice or text agent for customers

Answers, qualifies and books, with the compliance rules written into the code.

An interactive or 3D experience

WebGPU and real-time 3D on the web, from the engineer who shipped a PvP arena game.

A second opinion

An architecture, vendor or AI plan reviewed before you commit money to it.

Questions

How much does it cost to build an AI agent?

It depends on scope, so every build gets a fixed quote after a short discovery sprint. One automated workflow is a small project; a multi-agent system with voice, billing and monitoring is a larger one. The budget field below helps me answer quickly.

Do you only build AI projects?

No. Web apps, SaaS products, internal tools and 3D experiences are all on the table. AI goes in where it earns its place.

Who owns the code?

You do. Builds run on your own cloud account, and the source code, runbook and credentials are handed over at the end.

Can we start with just a call?

Yes. A consultation is the easiest way in: one or two hours on your idea or plan, with written next steps whether or not a build follows.

Start a build

Tell me what it should do. I reply with whether it is buildable, a rough cost and a timeline.

Want to talk first? Pick Consultation in the form.