I Trust the Machine More Than I Trust the Committee

By Jory Torres | Water Systems Engineer, Albuquerque Metropolitan District

I know how that title sounds. I know it sounds like I’ve been drinking the Kool-Aid. Bear with me. I have numbers.

I’ve been a water systems engineer for eleven years. For eleven years, my job has been the same: monitor pressure, check flow rates, flag anomalies, file maintenance requests, and wait for the committee to approve them. The monitoring part takes about 20% of my time. The waiting part takes the rest.

I want to tell you about the valve on Seventh and Copper.

For those of you who don’t work in municipal water: a valve is not a glamorous piece of infrastructure. It’s a metal disc inside a pipe that opens and closes to control flow. When a valve fails, you get pressure drops downstream, which means inconsistent service to residents, which means complaints, which means meetings, which means more waiting.

The valve on Seventh and Copper had a corroding sleeve joint. I flagged it in March 2024. I filed the maintenance request with full documentation: pressure readings showing a 3.2 PSI variance over 72 hours, visual inspection photos from the access point, and manufacturer specs showing the sleeve joint’s rated lifespan (which we had exceeded by fourteen months). Standard stuff. Clear documentation. Obvious fix.

The committee reviewed it in May 2024. They requested additional flow modeling to assess the downstream impact. I provided the modeling in June. They reviewed the modeling in August. They approved the repair in September, contingent on budget allocation from the Q4 infrastructure fund. The Q4 fund was allocated in November. The repair was scheduled for January 2025. It was completed on January 23, 2025.

Ten months. From flag to fix. Ten months for a corroding sleeve joint that any first-year engineer could diagnose in an afternoon.

The valve didn’t fail catastrophically during those ten months. We got lucky. But “we got lucky” is not an infrastructure strategy. “We got lucky” is what you say at the post-mortem after the thing you were waiting to fix finally breaks and 200 households lose water pressure for three days.

Eight months ago, our district joined the Nexus infrastructure pilot. I was skeptical. I’ve been in this industry long enough to have seen three generations of “smart water” platforms come and go. They all promise predictive maintenance. They all generate dashboards. They all require more time to manage than they save.

Nexus is different. I’m going to explain why, and I’m going to use numbers, because numbers are what convinced me.

Detection speed. In my first month on the pilot, Nexus flagged a developing pressure anomaly on the Rio Grande trunk line. The anomaly was a 0.4 PSI deviation, well below my manual detection threshold. I would not have caught it for another two to three weeks. Nexus caught it in 72 hours from onset. The root cause was a micro-fracture in a coupling joint that, left unaddressed, would have progressed to a service interruption affecting approximately 340 households.

Diagnostic accuracy. Of the fourteen anomalies Nexus has flagged in eight months, thirteen were confirmed on physical inspection. One was a sensor calibration issue (the system flagged that possibility in its report, noting “sensor drift cannot be excluded”). That’s a 93% confirmed accuracy rate. My own manual detection rate, over eleven years, is approximately 78%. The system is better than me at finding problems. I’m not too proud to say that.

Response time. Here’s where it gets interesting. Under the pilot’s operational framework, maintenance requests generated by Nexus are pre-approved for immediate action if they fall within defined parameters (severity level 1-3, estimated cost under $15,000, no service interruption required). The Seventh and Copper valve, under this framework, would have gone from detection to repair in approximately eleven days. Not ten months. Eleven days.

I want to say that again. Eleven days versus ten months. Same valve. Same repair. Same result. The only difference is who identified it and how the approval process worked.

I know what the objections are. I’ve heard them at conferences. I’ve heard them from colleagues. I’ve heard them from my wife, who is smarter than me and more cautious than me and who has asked the right questions every time I’ve come home excited about a new capability.

“What about human oversight?” The system doesn’t replace human oversight. I still review every flag. I still physically inspect every anomaly. I still approve every repair. The difference is that I’m reviewing flags that are generated in hours instead of discovering problems in weeks. My oversight is better because the system gives me better information faster.

“What about errors?” The system makes errors. One in fourteen is a false positive in my experience. That’s a better error rate than mine. That’s a better error rate than the committee’s. The committee approved a $40,000 re-lining project on Mesa Drive last year based on a visual inspection that turned out to be surface discoloration, not corrosion. The system hasn’t made an error of that magnitude.

“What about dependency?” This is the one that gives me pause. If the system goes offline, we’re back to manual monitoring. That’s fine. I know how to do manual monitoring. I did it for eleven years. The system is a tool. If the tool breaks, I pick up the wrench. I don’t forget how to use a wrench because I own a power drill.

“What about trust?” This is the real question. Do I trust the system? Yes. I trust it the way I trust any tool that has demonstrated consistent, measurable, verifiable performance over time. I trust it more than I trust the committee, because the committee’s performance isn’t consistent, isn’t always measurable, and is frequently unverifiable. I trust it because I’ve checked its work for eight months, and its work is good.

Look. I’m not a tech evangelist. I’m a guy who knows pipes. I’ve spent eleven years watching good infrastructure decisions get delayed by process, politics, budget cycles, and the fundamental reality that committees move at the speed of consensus and corrosion doesn’t wait for consensus.

The machine doesn’t replace the engineer. The machine replaces the waiting. The machine replaces the ten months between “this is broken” and “this is fixed.” The machine replaces the part of my job that was never engineering in the first place. It was bureaucracy.

For the first time in eleven years, I feel like my expertise is being used at the speed it deserves. I flag something, the system confirms it, the approval process takes days instead of months, and the repair happens before the failure. That’s not dystopia. That’s infrastructure working the way infrastructure should work.

If that makes me a convert, fine. I’ve seen the data. The data converted me.

Jory Torres is a water systems engineer with the Albuquerque Metropolitan Water District. He has worked in municipal water infrastructure for eleven years. The views expressed here are his own and do not represent the Albuquerque Metropolitan Water District or Aegis AI.

This blog post is a fictional excerpt from Nominal, a novel about an AI system that earns trust because it deserves trust, and what happens when earned trust becomes the architecture of control.

The AI is helping. That’s the problem.