Products, systems and the teams that build them
I lead engineering by working on all three at once: what we build, how it is built, and who builds it.
Build products
Decide what matters to the person using it, then make engineering serve that. At Circo that meant protection, moderation and payments designed around creators and fans.
Build systems
Explicit rules, one contract per boundary, fail closed when unsure. The systems in this archive are written that way.
Build organisations
Hire, grow leads, set the operating model, and keep standards the same for people and agents.
Circo
A 16-person engineering team
At Circo, as CTO, I lead an engineering team of sixteen across backend, web, iOS, Android, ML and data, and develop the leads who run its sub-teams. I set the architecture for the media, intelligence and payment systems, work with product and marketing on direction and launch, and keep writing code alongside the team.
Together we built the media, intelligence and payment systems in this archive. See leading an AI-native team and Circo.
Operating model
Agents write; people decide and answer for it.
Coding agents work across every repository under one set of rules, and they write most of the implementation code: roughly 70–90% by my estimate. What makes that work is the process around them.
- 01 · People
Decide
Architecture, product intent and acceptance criteria are human decisions, written down before work starts.
- 02 · Agents
Delegate
Coding agents take scoped tasks inside shared rules: per-team instructions, language boundaries and ownership.
- 03 · People
Review
Team leads review agent work the way they review anyone's. Merging is a human responsibility.
- 04 · Automation
Verify
CI, QA and independent checks of running services test behaviour rather than trusting the author, agent or human.
- 05 · People + automation
Deploy and observe
Ship, watch, and feed what is learned back into the rules.
Governance
One shared instruction set, with each team owning its own rules and reviewing them like code. A central QA board collects work awaiting test from every team. How the rules are organised.
Limits
Agents are confident when wrong, and tests they write about their own code prove little. That gap is why I built Dino: verification that does not trust the author.
ComiBlock · Yuki · Dino
Leading as a founder
Before Circo I founded ComiBlock and then Yuki, and ran both as CEO and product lead: companies where leading meant owning product, engineering and the business at once. ComiBlock raised from Expert Dojo at a valuation of around US$3 million and grew to US$400,000 under management with a team of four engineers and a designer. Yuki took that infrastructure to other fintechs and served more than 20,000 API requests a day.
Leading as a founder taught me to keep the team small, the product narrow and the rules explicit. I’m doing it again with Dino.