The Ops Team’s AI Layer

Most of what gets called “AI in operations” is a chatbot bolted onto a help page. I’ve come at it from a different angle: I treat AI as a layer that does the actual work — the drafting, the auditing, the reconciling — and leaves me as the reviewer instead of the typist.

People-ops is full of work that’s too structured to deserve doing by hand, but too irregular to hand off to ordinary automation. Filling one employment contract means a form, the right template out of twenty-plus jurisdictions, and forty careful minutes. A census audit means a week of spreadsheet archaeology. Neither is worth spinning up a software project for — and both come around constantly.

A contract-generation skill that reads the intake form, finds the right jurisdiction template, applies the entity and salary logic, and hands back a draft that’s ready for legal in minutes. Every edge case it knows how to handle is one I hit myself, by hand, first.

A census-audit skill that takes the matching engine from the HRIS migration — the tiered email-then-name matching, the multi-pass gap analysis, the formatted reporting — and folds the whole thing into a single prompt.

Both of them run on live payroll, in production. That’s the bar I care about — not a demo I can show off, but something the team would feel the absence of if it disappeared.

There’s a habit underneath all of it: I write specs and playbooks precise enough that an AI can actually run them — which turns out to be the same muscle as writing payroll documentation precise enough that an auditor will trust it. Each one keeps making me better at the other.

Contract turnaround went from a queue measured in days to a review measured in minutes, and census audits now run whenever I need them. The change I value most is smaller: every new payroll problem gets a second question after “how do I fix this?” — namely, “how do I make this a one-prompt problem next time?”