Where AI genuinely helps a public works team (and where it doesn’t)

AI shows up on nearly every asset management vendor’s homepage. It shows up in almost none of their actual software. There’s a gap between the roadmap slide in the sales deck and the screen your team logs into Monday morning — and that gap is where most municipal AI claims live.

The roadmap problem

Ask a vendor to show you the AI feature live, today, in production — not a mockup, not “coming next quarter.” For most legacy platforms, that request stops the demo cold. AI in municipal software has largely meant a slide, not a shipped capability. That’s not a knock on any one vendor; it’s a pattern across an industry that’s been slow to rebuild around it.

What shipped AI actually looks like on the admin side

Standing up a new city in a legacy CMMS usually means weeks of manual configuration — work order types, inspection checklists, notification rules, asset categories — built one screen at a time, often by a consultant billing by the hour. Shipped AI changes that starting point: it can read what your city already has — your GIS, your existing documentation — and propose the configuration instead of making your team build it from a blank screen.

What shipped AI looks like for the people doing the work

For end users, the test is simpler: can you ask it a real question and get a real answer? “How many work orders are overdue this month, and which crews?” shouldn’t require a report-builder or a call to IT — it should be a sentence, answered with the actual records behind it, not a canned response. The same is true for building a dashboard, drafting a schedule, or finding a specific asset’s history. If the answer requires clicking through six menus, the AI isn’t doing the work — it’s decoration.

Where AI doesn’t help — and shouldn’t pretend to

AI shouldn’t invent inspection data, guess at a capital plan, or replace the judgment of someone who’s walked the asset. The honest version of AI in public works software augments the people doing the work — it doesn’t substitute for them, and any vendor implying otherwise is overselling. A good assistant tells you what it found and shows you the record it found it in. It doesn’t ask you to just trust it.

The real buying question isn’t “does it have AI” — nearly every vendor will say yes. It’s whether you can open the product today and ask it something. For most of the market, that’s still on the roadmap. It shouldn’t have to be.