The note-taking button
In most offices I visit, AI means someone pressing a summary button after a meeting. Useful, but small.
The real gains come when AI changes how the work flows, not when it decorates the old flow.
Order before intelligence
An agent is only as good as what it can find. Inconsistent folder names and files scattered across personal drives will defeat any model.
So the first project is usually unglamorous: a standard folder structure, clear permissions and a reliable backup that does not depend on a single vendor's cloud.
Then a knowledge layer
Once the archive is orderly, a knowledge graph can connect projects, documents and decisions, so people find what they need even when they misspell it.
That is where search stops being a chore.
Pick one heavy workflow
Choose a task that is repetitive and expensive, such as reading tender documents. Build an agent for it, keep a person checking its output, and measure the time it saves.
One workflow done well teaches an office more than ten tools installed.
Pilot small, own the result
Start with eight to ten people, not the whole office. Give the project an owner and protected hours.
Whatever you build stays the office's own asset, with permissions and security designed in from the start. Look for quick, low-cost wins first; they buy the patience for the rest.
How we use it ourselves
At Realization, AI is how a small team stays fast: research, documentation, coordination and follow-up.
It is an internal edge, not the product. The same discipline works in any professional practice.