It started one early Saturday AM . . . with a watermelon.
Which, I admit, is not usually how I begin a discussion about Public Safety technology. There were no acronyms involved. No standards document. No federal rulemaking. No vendor slide deck with seventeen arrows, four clouds, and a glowing icon labeled “AI-powered situational intelligence.”
Just a 12-inch round seedless watermelon sitting in my fridge after Friday’s AI-assisted shopping session, and an Instacart delivery – BTW Tiffany, 5-Stars for the produce selections!
The weekend weather was forecast to be spectacular. That meant lunch and dinner on the grill, eating outside, enjoying the sunshine, trying to get some color, and pretending for a few hours that email, meetings, policy debates, and whatever new acronym someone invented last week did not exist.
Somewhere in that plan, watermelon made a lot of sense.
And honestly, over the last 64 and a half years of my life, I have cut up my share of watermelons. Especially in the summer. I didn’t remotely consider this an unsolved operational problem requiring artificial intelligence, advanced prompting, or computational assistance to get through to solve.
It is a watermelon.
- You cut it.
- You eat it.
In theory, this should not require a next-generation architecture.
But lately, because I have been spending a lot of time around AI best practices, agentic workflows, prompt design, hallucination detection, tool usage, and building practical AI-assisted systems, I have started to use AI differently.
Not as a magic answer machine.
Not as a slightly more confident version of “Hey Google, what about blah blah blah?”
And definitely not as the digital oracle of truth, wisdom, and kitchen knife safety.
Instead, I have been using AI as a structured assistant. A tool that works best when the task is clearly defined, the guardrails are understood, and the expected output is described in a way that gives the model a fighting chance to be useful.
That may sound like overkill for cutting up a watermelon.
But that was exactly the point.
This was a low-risk task. The failure mode was not catastrophic. If I totally botched the job, I was out less than ten bucks and could probably replace the watermelon before lunch. That is a very different risk profile than testing a new operational process during a live emergency.
So I decided to ask AI a real question, but not in a lazy way.
I did not ask, “How do I cut a watermelon?”
That would probably have produced a generic list copied from the collective memory of every lifestyle blog ever written.
Instead, I gave it a role, a task, a condition, and an expected format.
I said:
You are a chef in a commercial restaurant. Your job is to cut up a 12-inch round watermelon. I want you to use the most efficient way possible and explain in a step-by-step manner the exact procedure to extract the most out of the watermelon. Provide that to me in a structured list of instructions that I can follow step by step.
That prompt mattered.
The answer was not just “slice it into wedges.”
It walked through a commercial-kitchen process. Wash the outside. Stabilize the cutting board. Cut flat ends. Stand the watermelon upright. Remove the rind in vertical strips. Trim the white pith carefully. Cut the peeled fruit into slabs, then sticks, then cubes. Save the clean trimmings for juice or secondary use.
In other words, it gave me a procedure.
And here is the annoying part.
It worked.
I got less waste. I had cleaner cuts. I had usable leftovers that I shrink-wrapped for later. And I ended up with a nice bowl of restaurant-style fresh watermelon cubes chilling in the refrigerator for lunch.
Was it life-changing?
No.
Was it useful?
Absolutely.
Did it make me slightly irritated that AI gave better watermelon instructions than 65 years of “yeah, I know how to do this”?
Also, a big yes.
But that is where this becomes a matter of Public Safety.
Because the value was not really in the watermelon.
The value was in the process.
I had a task. I defined the role. I explained the objective. I identified the constraints. I asked for a structured output. I used the result in a low-risk environment. Then I evaluated whether it actually worked.
That is AI literacy.
Not hype.
Not magic.
Not “the machine knows everything.”
Just a practical example of using a tool within clear boundaries to improve the outcome of a task. And that is exactly where Public Safety needs to spend more time.
Too often, we talk about AI like there are only two options. Either it is going to save the world, answer every call, write every report, dispatch every unit, detect every threat, and bring coffee to the night shift.
Or it is a dangerous hallucinating robot that should be locked in a digital closet until the attorneys figure out who to blame.
Reality, as usual, is more boring and more useful.
AI is a tool.
A powerful tool, yes. A tool that requires guardrails, governance, validation, policy, and adult supervision. But still a tool.
And the best place to learn how to use a powerful tool is not during the highest-risk moment of your day. Public Safety should not be learning AI for the first time during the emergency; that is like deciding to read the fire extinguisher instructions after the curtains are already on fire.
Start with low-risk use cases. Summarize meeting notes. Draft training outlines. Compare policy language. Build checklists. Create tabletop scenarios. Improve public education content. Help structure after-action reviews. Generate healthier snack ideas for the midnight shift, apparently.
The point is not to replace human judgment.
The point is to strengthen the human process before the stakes become unforgiving.
There is also a major difference between asking a vague question and crafting a useful prompt.
“Hey Google, what about . . .?” is not a workflow.
It is a digital shrug.
A good prompt is closer to a good dispatch protocol. It gives context. It establishes the role. It narrows the task. It defines the output. It sets expectations. It reduces ambiguity.
That does not guarantee perfection. Nothing does.
But it improves the odds.
And in Public Safety, improving the odds is kind of the job.
This is especially important because AI can sound confident even when it is wrong. That is not a small issue. That is the trust gap. Just as a caller can sound real, a location can look right, and a vendor claim can arrive wrapped in polished language, an AI answer can be beautifully formatted and still be incorrect.
- So we verify.
- We test.
- We start small.
- We build mental muscle memory before consequences arrive.
That is why a watermelon is not a silly example. It is a safe example.
- No watermelons were hurt in the process.
- No emergency response was delayed. No dispatcher was overloaded with bad data.
- No PSAP policy was violated or rewritten based on unverified output.
The worst-case scenario was me standing in the kitchen surrounded by hacked-up fruit, quietly blaming the artificial intelligence when the real problem was that I did not follow directions.
Which, to be fair, is not exactly new behavior for me. I’ve been known to mess things up because I didn’t follow directions, even when I am the one who gave them.
That may be a personal operating flaw, but at least it is well documented. And that is another lesson. AI does not remove the human from the process. It makes the human more important.
The person still has to ask the right question. The person still has to evaluate the answer. The person still has to apply judgment. The person still has to know when the output is useful, when it is nonsense, and when it needs a second look.
That is true in the kitchen.
It is much more true in the PSAP / ECC.
And yes, somehow this is still about Public Safety. Despite all the technology, the acronyms, the platforms, the dashboards, the AI engines, and the blinking things vendors love to put in PowerPoint decks, the most important asset in emergency response is still the human under the headset.
That asset needs maintenance.
It needs rest.
It needs support.
And occasionally, it needs something better than whatever fossilized snack came out of the vending machine at 2:00 in the morning on a Thursday, when the snack guy isn’t expected to show up until next Wednesday.
So as summer rolls in, maybe a Ziploc bag of cold watermelon cubes is not just a snack. Consider it a tiny operational wellness plan with seeds removed.
Stay well. Consider something easy to do, like thinking about a healthier snack. And if you are not sure what better choices look like, I happen to know a very large language model that can help with that too.
Thanks for hanging out with me for a little while. I’d love to hear your comments and appreciate you following me on social media. Remember, my weekly episodes of “TiPS: Today in Public Safety” are published every Monday, Wednesday, and Friday, with links on most social media platforms and at 911TiPS.com. If you’re in public safety, thanks for all you do every single day. Until next time.
Mark Fletcher, ENP • Founding Principal • fletch911.com