Building conversational agents and running engineering at Norma. Trying to master the whole AI stack. Writing about friendship and dating.
How I run teams living
As an engineering leader, my job has three components: build the thing (sometimes we know what the thing is, sometimes we don't), tell leadership what we learned making it, and develop my team.
I'm an advocate of bleeding edge organizational AI adoption. In development practices, management review, and everything in between.
Read on for how I use AI, what I expect from my directs, and how to best work with me.
Code reviews in the age of AI living
To code review or not to code review. That is NOT the question.
Once agents multiply everyone's throughput, the scarce resource in an engineering org stops being engineering time and becomes review attention.
Code reviews were never one activity. Style belongs to linters, bug-catching to LLM review agents, correctness to test suites. The human functions that remain are design and alignment, so I run design review by default and code review by trust tier.
My personal agent pipeline 2025–26
My personal AI coding stack primarily consists of an OpenClaw instance (running on the latest GPT model) and Claude Code.
I have a combination of heartbeat-powered automations and cron jobs that form an agent loop as well as a myriad of personal drafted skills for hands on keyboard prompting.
Together we've built a personal Wikipedia over my notes vault, a nightly pipeline for analyzing journal entries, synthetic users who review my products, a pipeline for generating spaced-repetition cards from my recent reading, a bot that books my pickleball slots the moment they open, and much more.
As much as loops are the talk of the town, I keep finding that agent capacity is rarely the limit and quality input is truly scarce.
Books as code 2026
I wanted to know whether AI agents could help me speed up the design process to create a high-craft physical book. I knew it was possible to generate text, but could we create layouts, covers, and coherent design language?
The answer was yes! For my girlfriend's birthday, I authored the story of our relationship told through drinks.
AvaCloud, enterprise blockchain as a service 2022–24
Ava Labs had invented subnets, a way for anyone to run their own blockchain, but deploying a hello world demo took weeks of specialist work. I built Avalanche CLI to get that down to five minutes. Then, watching people use it, I changed my mind about what we were actually selling. The enterprises who most wanted their own chain were the least equipped to run their own infrastructure. They wanted to point, click, and buy.
So I photoshopped an EC2-style dashboard for subnets and pitched it at a company meetup in New York, where I was mostly told to stay in my lane. A few months later I pulled a team together at an internal hackathon and won the top prize. Leadership gave us the green light to build it for real, so I started out as the sole engineer while I built an actual team around it.
AvaCloud grew into one of the company's three main lines of business, and the client wall now includes FIFA, J.P. Morgan, Deloitte, and Konami: companies that want their own private blockchain without becoming infrastructure operators. I led the engineering effort from photoshop mockup until I left, by which point the team had seven engineers plus over twenty headcount across product, QA, marketing, and solutions.
