I build the infrastructure that makes AI agents reliable in production.
Two decades moving up the stack: embedded firmware and mobile apps, then carrier-grade networks, then cloud platforms, and now multi-agent systems. Career stops include Apple, AT&T, Verizon and Sam's Club.
Three ways in
-
/crash-test/
CRASH/TEST
Ten real AI agent disasters, replayed with and without safety controls. Paste your own agent configuration and see which of the ten it would stop.
-
/mind/
The Ship Mind
An AI that has absorbed my career record. Ask it anything about the work; it answers only from the facts on file.
-
/play/
Solar Mandate
A daily three-minute strategy game. Wire up the solar system's internet before the light runs out.
Products I run, with prices, are on the tools page. Hiring managers can go straight to the hiring page.
The network so far
Every visit, game, crash test and question leaves one anonymous signal on a shared map. The more people pass through, and the more of them arrive through each other's links, the more the map fills in.
- visited
- flew the Mandate
- ran a crash test
- questioned the Mind
- a line joins a visitor to whoever's link brought them
Today's Mandate
- Scores appear here as people fly today's Solar Mandate. Everyone gets the same solar system each day.
Scores are reported by the game and are not verified. Names are generated, never typed.
How it grows
- Arrive. Your browser gets a callsign and a dot on the map.
- Do something. Fly the Mandate, run a crash test or question the Mind, and your dot changes colour.
- Pass it on. Anyone who arrives through your link is wired to your dot.
- Everyone gains. Every petabyte flown counts toward opening the next ring of the map.
- Signals today
- 0
- Visitors, all time
- 0
- Mandates flown
- 0
- Crash tests run
- 0
- Mind interrogations
- 0
- Petabytes, together
- 0
You are .
Up the stack
-
2024 to now
Agents and software factories
AI agent architect and software factory creator. Independent, Nubla AI and open work.
Multi-agent orchestration, deterministic supervisors that catch model errors, context engineering, and interpretability tools. The aim is infrastructure that lets one engineer do the work of a team without constant supervision.
-
2022 to 2023
Cloud at consumer scale
Cloud architect and engineer, Apple.
Cloud services and infrastructure for products where uptime and polish are not negotiable: latency budgets, failure domains, operational simplicity.
-
2017 to 2021
Cloud and machine learning
Principal software engineer, Sam's Club.
Brought production machine learning into core retail workflows, with the monitoring, versioning, reliability and cost control that takes.
-
2007 to 2009
Carrier-grade applications and services
Mobile and web applications architect lead, AT&T. SOA designer and architect, Verizon.
Platforms for carrier-scale traffic, where uptime is counted in nines and change is deliberate. The habits of thinking in failure domains started here.
-
Before that
Embedded firmware and mobile applications
Where the climb began.
Newest layer on top. Dates are the ones on file in the career record; the Ship Mind can tell you more about each.
Open work
- factory-gate
Policy guardrails that verify an agent's state transitions are safe before code reaches a target branch.
- circuit-x
CircuitX: mechanistic interpretability tooling for reverse-engineering neural weights and code models.
- deterministic-agent-supervisor
A deterministic supervisor that keeps agent workflows from acting on model errors.
- sliding-window-trim-engine
Sliding-window token management for long agent runs.
How I work
- Design for failure, then make failure boring.
- Latency is a feature. So is clarity.
- Leave systems, and people, better than you found them.
- The highest leverage is infrastructure that works without you.
- Reliability and interpretability are non-negotiable at scale.
What I work with
- Domains
- Multi-agent orchestration and software factories. AI harness and context engineering. Edge and distributed systems. Production ML and reliability. Interpretability and alignment tooling. Networking and security architecture. Observability and cost engineering.
- Platforms
- Cloudflare (Workers, Pages, R2, D1, KV, Zero Trust), AWS, GCP, Azure, Kubernetes, Docker, GitHub Actions.
- Languages
- Python, TypeScript, Go, Swift, Java, Rust, JavaScript, C#.