Homework, Done Well

A client asked for an API, and they brought the spec. Red Folder Research does insurance-policy research for personal-injury law firms. The spec had software startup ambitions.

OAuth 2.0. Two grant types: authorization code, and resource-owner password. A role for every lawyer type at every firm. Auto-scaling, horizontally. A login screen, and a multi-tenant design and product behind it. The studying was done well, this is all industry standard. The document knew what a proper API looks like.

Their use was way smaller. Humans submit cases. Humans manage steps. Humans do the nitty gritty of the work. Their "factory" would still run on humans. Jira was their interface, and it'd stay so - they already live there.

So I cut the exotic accounts, one layer was more than enough. Service-to-service tokens plus Jira login. Access at the firm level, it's business to business product, their interns don't need to look behind the scenes.

Horizontal scaling isn't hard, but it's expensive to design around. Their business would need to 1000x for a 32tb.480xlarge (AWS) to struggle. Vertical works out of the box, so that was the call.

Clients were filling Jira forms. Researchers were waiting on a pile of Atlassian automations to stand in a queue over a singular case. They're problems solved via technology, yes, but they're not technology-problems. The supplied spec was talking about technology problems from the point of view of the tech industry. Behind the spec, the ask was to avoid another hire, and the machines weren't to blame. Optimize for the problem, not for the wider industry.

A spec can be correct and still be the wrong job. I acknowledged the homework. I did not build it.

The rest of that build: https://tinthe.dev/case-study-red-folder