The agent already did the busywork. The process did not notice.

The agent already did the busywork. The process did not notice.

An ops person can now research buildings, pull utility rates, cross-check providers, and draft the client note in one sitting. That is happening. Most workflows still assume a human does every fetch.

AI is good at patterns at scale, boring classification, and first-pass research. People are good at what matters, what the data missed, and what the training set never saw. Design for that split.

Three rules that do not need a transformation office

Find the steps that look the same every time: entry, routing, status checks, first research. Invoice layouts vary, but the job is the same. Read the line items, match the PO, send what is 95% sure, park the rest for a person. That is the pattern.

Keep a human on anything that commits the company. A youth program scored attendance risk in seconds. A coordinator still decided who got a call. Drafts and scores first. Sign-off before it goes out. Tighten review only after you trust it.

Match the tool to the data you actually have. Forty thousand thin rows may not train a custom model. ChatGPT or Claude on the question you have today might. Start general. Specialize when the pile is big enough and clean enough.

Give people the exceptions, not the keystrokes

Ops: vision for the obvious defects, inspectors on the maybe. Finance: extract and code, people on the misses. HR: early signals, managers on the conversation. Marketing: research and a draft, then the brand. Support: sentiment to route, veterans on the hard tickets.

People leave when the day is copy-paste. Map one process. Automate the repeats. Review the output. Measure. Then the next one. Assign the work to whoever is better at it. That is the whole idea.