Building Internal Tools with AI Assistance

The short version

AI coding assistants have genuinely lowered the floor. Somebody who understands the department's work, but has never shipped software, can now produce a working internal tool in an afternoon. The work that suits this well is administrative: workflow that is currently a spreadsheet and three emails, equipment tracking, internal forms, dashboards, data cleanup, small automations, prototypes to show what you mean, and training tools. What has not changed is responsibility for the result. AI-generated code is not automatically secure and not automatically correct. It is confident in both cases. It will happily write something that works on your test data and fails on the real thing, or that stores a password in a way no one would choose deliberately. So treat generated code the way you would treat work from a capable but unfamiliar contractor: useful, fast, and requiring review by someone who understands what it is touching before it goes anywhere near real data.