A six-part build log of one live system, Bullpen, a paper-trading league on real market data, used to test which microservice and distributed-systems decisions still hold in 2026, cost included. It records my take at this moment: the platforms under it change monthly, and some of it will age.
I am building a real distributed system in public and making every split earn its place with a check, a measurement or a bill. This is a dated snapshot of how I would architect software in 2026, not a timeless rulebook.
Companion project
GitHub
Source Code and plan for this project
The parts
- 01When to Use Microservices in 2026: Scaling Alone Is No Longer the Reason
On the day the Dow posted a then-record gain, Robinhood's customers could not trade from the open to the close. Splitting out the hot part so it scales alone was the old answer; a serverless platform now does that per function. I open this series with Bullpen, a paper-trading league on live market data, and argue when to use microservices in 2026: only for a reason I can name, with a check, a measurement or a bill behind it.
EngineeringOct 4 · 12 min read - 02Architecture as Code Has Three Jobs: Describe, Govern, Remember
In Bullpen's first week, the planned architecture model still named a stock provider I had turned down, and a footer change and a test followed it. Nothing failed. I use that incident and two others to argue that architecture as code has three jobs: describe the system, govern it, and remember why it looks the way it does. Then I follow the first real change after the fixes, product analytics, through all three: a decision record before the model could name it, built and planned architecture that disagreed, and a rule that composes the two instead of editing either.
EngineeringOct 11 · 11 min read - 03Database per Service Is Free Now. That Is Not a Reason to Use It
A free tier that doubles changes the inputs to a decision, not the decision. In week one I predicted Bullpen would split prices and history into two services with two databases. Then Neon made database per service look free, and I had to decide whether that changed anything. This is the decision history: five data and deployment decisions, each with its conditions, the options I weighed, why one won, and what would make me reverse it.
EngineeringOct 19 · 28 min read - 04Not yet publishedNext
- 05Not yet publishedPlanned
- 06Not yet publishedPlanned
3 parts to go. New parts land on Mondays, 9am Pacific — leave an address and I'll send each one the day it ships. Nothing else.