When the winter storms hit my city, Orting, Washington, I did not ask my team how to plow the roads. I asked them one question. How will we have failed?

Stand at the end of the disaster and look back at it. Assume it already went wrong, then trace the wreckage to its cause. Snow on the road was never the failure. An elderly resident freezing in a dark house was the failure. A man who could not charge his breathing machine was the failure. A family that never heard the evacuation order was the failure.

Once we named those specific failures, the work stopped being a guess. We found those people first when the predictable power outages came. We carried hot soup door to door. We knocked on the doors that mattered. We beat the storm before it arrived.

The weather was never the thing to fear. Yet every time the news called, they asked the same question. Are you ready for the storm? It is the wrong question. We are conditioned to treat what we cannot control as failure. It is not.

The snow was going to fall whether I was ready or not. The failure I could actually own was the resident we did not reach in time. You cannot fight the storm. You accept that it is coming, and you pour everything you have into the part you control.

That question ran my whole city.

It ran a budget through a fractured council. It ran a wastewater permit when a warning letter from the Department of Ecology collided with an atmospheric river bearing down on the plant. It ran a flood when the river overtopped our levees and the national cameras showed up wanting a disaster we had spent a decade engineering away.

Every time, I asked the same thing first. How will we have failed? Then I worked backward until the failure had nowhere left to hide.

That question also runs the problem sitting in your building right now, the one most councils have not named yet. That problem is artificial intelligence.

Start here, because this is the part most leaders skip. Your staff already use these tools. This is not a line item for next year's strategic plan. It happens this morning, on your network, with your residents' data.

Somewhere in your building right now, a clerk pastes a resident's email into ChatGPT to draft a faster reply. The name. The street address. The complaint about a neighbor's fence. She is not reckless. She is buried. She has forty tasks and thirty minutes before the counter opens, and the tool clears one of them.

Down the hall, a planner drops a zoning question into the same window. A finance analyst asks it to summarize a vendor's contract. A code officer uses it to soften a violation letter. None of them act in bad faith. They do what capable, overworked people always do. They found a tool that saves ten minutes, and no one ever told them where the line sits, because you have not drawn one.  And almost no one has drawn it.

In ICMA's own survey, only about a third of local governments reported a defined AI policy, and roughly another third were still building one. Most jurisdictions are taking a wait-and-see approach. Understand what wait-and-see actually means. It is not neutral. It is the decision to let every employee with a browser tab set the policy for you, one keystroke at a time.

Here is why this slips past good leaders. The danger does not arrive as a siren. It arrives as convenience.

I have written before about the gap between the pothole and the pixel. The pothole is the real, messy problem on the ground. The pixel is the clean green checkmark on the slide that tells you the problem is handled. A shielded executive manages the pixel and never touches the pothole. AI widens that gap and paves right over it.

A tool that writes a confident, polished paragraph in four seconds produces a beautiful pixel. It looks finished. It looks correct. Underneath, the AI may have invented a code section that does not exist, leaked a resident's file to a server you do not control, or quietly created a public record you will hand a reporter next month.

And the bubble only gets thicker.

Every executive already fights to get the unvarnished truth past staff who would rather hand you good news. AI hands those same staff a machine that manufactures good news on demand. A report that used to take a week to dress up now takes a single prompt. The polish keeps improving while the ground truth gets harder to reach.

If you do not build your own channels to the actual pothole, you will run your city on pixels that look flawless and describe a place that does not exist.

Join me for a Free Workshop - click image below

Here is where a lot of elected leaders freeze. You are not a coder. You are not a lawyer. You will not draft the ordinance yourself, and you will not build the AI itself... and you should not.

Your staff and your council write the actual policy.

Your work is different, and only you can do it. You decide the question gets asked that guide all of it. You put it on the agenda before a crisis puts it there for you. You set the direction, then you get out of the way and let capable people build (policy and technology). And you own the risk when the tool gets something wrong, because "the software did it" is not a defense a single resident will accept, and it is not one you would accept from anyone who works for you.

So do the one thing the chair exists for. Ask how your city will have failed. Then send your people to work backward from the answers.

Before the specific failures, name the biggest one. You fail the moment you decide to write AI off or ban it outright. Both moves feel responsible. Both are foolhardy. Banning AI in your organization today is like banning email in 1995 because it looked like a nuisance.

Your staff will use it anyway, on their phones, off your network, with none of your rules and none of your protection. As I learned in every storm, you cannot fight the weather.

You accept that it is here, and you govern the part you control!

Everyone knows the post-mortem. A project fails. An initiative collapses. Leadership gathers in a sterile room to point fingers and dissect the corpse. By the time you run a post-mortem, the money is gone, the trust is spent, and the headline already ran.

The pre-mortem flips the order.

You gather the same people, and you tell them the AI rollout has already blown up. Six months from now. The local paper has the story. The council is furious. A resident's private file is loose on the internet, and a reporter has a copy. Now the room's job is not to defend the plan. It is to explain, in detail, what went wrong.

Ask your clerk how the record got out. Ask your IT lead how the data left the building. Ask your department head which decision the AI made that a human should have made. Name each failure out loud. Then, and only then, write the rule that would have stopped it.

I have run this drill when the stakes were not hypothetical. One year a warning letter arrived from the Department of Ecology about the solids lagoon at our wastewater plant.

On its own, a routine permit problem. Then the forecast turned, and an atmospheric river started tracking straight at us. If that lagoon overflowed during the flooding, the ecological, legal, and financial fallout would have gutted the city.

A reactive leader panics and starts assigning blame. Instead, I sat the team down and assumed our primary plan had already failed. Then we worked backward through the logistical failure points, one at a time.

If our main dewatering bid falls through, what is the backup? Geotextile eco tubes. Can we physically receive tubes that large in time? Can the supplier deliver on a weekend? Do we hold the free capital to pay for them now? Do I need emergency authorization from the council tonight, or do statutory emergency powers cover it? At what exact moment do I declare a state of emergency?

We answered every one of those questions before the water arrived. The storm hit. The tubes held. The facility held. We moved forward with mitigation – we did have an emergency, but a manageable one.  We stayed in compliance, and the crisis that should have made the news simply did not happen.

Run that same drill on AI. Assume the rollout already failed. Then ask the logistical questions in advance. Where did the data go? Who approved the tool? Which record did we lose? Which decision did the machine make that a human should have made? You answer those on a quiet Tuesday, in a conference room, so you never have to answer them later in front of a camera. Here are the failures worth naming, and the rules that answer them.

You fail when private data walks out the door. So draw the data line. Decide, in plain language, what your people may never paste into a public AI tool. A resident's private information. Anything you would withhold from a records request. Anything tied up in a lawsuit or an active investigation. Anything a state or federal rule already protects. This is the single clause that prevents most of the real harm, and you do not need a study committee to reach it.

Your clerk, your attorney, and your IT lead can draft it in an afternoon, and your council can adopt it this month. Write it so a new hire understands it on day one, because a new hire will test it on day one.

You fail when the record vanishes. Your employee hits enter, and in most states that prompt and that answer become a public record the instant they exist. If your retention schedule and your disclosure rules do not mention AI, they went out of date the day your staff logged in.

Picture the request that lands next spring. A reporter, or an opponent, asks for every AI prompt your staff entered about a controversial project. Can you produce them? Can you defend what they say? If you cannot answer that today, your clerk should not be the one to discover it under a deadline with an attorney on the phone.   Is your assertion that prompts are transitory?  That they are essentially like Google search terms?  Make sure you’ve assessed that with legal before you’re defending it in court.

Update the schedule now, while it is boring, so it is not an emergency later.

You fail when no one owns the mistake. "The AI got it wrong" will not calm an angry resident, and it should not calm you. A tool that is right ninety-five times will be wrong the ninety-sixth, and the ninety-sixth is the one that ends up in front of your council. So name the human.

Name the person who reviews AI-assisted work before it leaves the building.

Name the decisions AI never makes on its own. A benefits determination. A code enforcement action. A hiring screen. Anything that changes the course of a person's life.

The machine can draft. A person decides. Put that in writing before the wrong draft goes out under your city's name.

You fail when the vendor plays your council. A vendor will demo a tool that promises to answer your permit questions, write your resident emails, and cut your backlog in half. The demo will run smooth. The council member who read one article will love it. The pressure to buy will land on you. So ask the three questions the salesperson hopes you skip.

·      Where does our data go, and who else can see it?

·      Do you train your models on what we put in?

·      Show me how you tested this for bias, in writing, with results.

A vendor who answers cleanly has earned a real look. A vendor who deflects has told you everything you need to know.

The one pushing hardest on your council has earned the most scrutiny, not the least.

You fail when you bet the whole city on an untested tool. So do not. You did not rebuild the whole levee system in one season, and you do not roll AI across every department because one demo impressed a committee.

Pick one low-risk job. Summarizing public meeting minutes. Drafting routine correspondence a human then signs. Put a hard stop on it. Measure it for ninety days against the thing it replaced. Then let what you actually saw decide what scales next. Not the sales deck. Not the loudest voice in the room. Build your AI muscle on purpose, in a corner of the organization where a mistake costs you a week instead of a lawsuit.

Do these pre-mortem planning actions for a year and something will change in your building. I guarantee it. Your people stop waiting for permission and start seeing the pothole under the pixel on their own. They flag the risk before it becomes a crisis. They tell you the hard thing early, while you can still act on it, instead of polishing it into a clean report that hides the truth until it is too late to fix.

That is the whole game. In Orting, our biggest wins never made the news, because a win in this work is a disaster that never happened. The lahar that never reached a home. The flood that never took a life. The lawsuit that never got filed. The headline that never ran. Nobody hands you a ribbon for the fire that did not start. You do the work anyway, because keeping your residents safe and your city out of the paper is the job you signed up for, whatever the title on the door says.

Today YOU can prevent the next fire – the AI fire.

Your AI headline is still unwritten. You can keep waiting. Wait for the state to hand down guidance. Wait for a model ordinance from a city that looks nothing like yours. Wait for a consultant with a fifteen-thousand-dollar binder and a four-month timeline.

However, every week you wait, your employees write your AI policy for you, one keystroke at a time, and you will not like what it says when you finally read it.

Or you can put one question to your people before your next budget cycle. Assume it has already gone wrong.

How will we have failed? Name the failures. Write the rules that stop them. Then go make sure you don't.

That is not a technology strategy. It is leadership, pointed at a new kind of storm. You already know how to do this. You just have to ask the question before the snow starts falling.

I’m now filling reservations for my 4 week course on AI in Local Government.
If you’re not sure how to approach what might be the biggest topic to ever hit local government, this is how you get ahead of the wave before it crashes over you.