In a quiet shift that most developers missed, the same infrastructure that runs your test suites is now watching municipal water mains for leaks. A growing number of cities are folding continuous integration runners into their water-loss programs, and the rebates that once covered hardware sensors now cover the compute credits that keep those sensors honest. For engineers who spent years tuning YAML pipelines and fighting flaky tests, the skill set suddenly has a new buyer: the public works department.

The Rebate That Rewired the Water Utility

When a mid-sized city in the Pacific Northwest updated its water conservation rebate catalog in late 2024, the line item for leak-detection hardware was no surprise. What raised eyebrows was the addition of a line for "continuous monitoring software and associated compute." That wording, buried in a 40-page PDF, meant residents and small businesses could now be reimbursed for the cloud credits that run scheduled leak checks on their properties.

Municipal rebates have historically been a capital-expenditure game: you buy a smart meter, the city pays you back. But the new framing treats software as a service, an operating expense that can be metered and rebated. It is a small procedural change with large consequences. Procurement officers who once wrote checks for physical boxes now need to understand usage-based billing, data retention, and API rate limits.

The result is that CI runners, the same virtual machines that execute your build and test jobs, are being pointed at water pressure sensors instead of unit tests. A cron job triggers a pipeline that pulls flow data, runs anomaly detection, and files a work order when a leak is suspected. The infrastructure is identical to what powers a modern software delivery pipeline; only the artifacts have changed.

For engineers, this is a signal. The money that funds public works is starting to flow toward the tools they already know. It is not a trickle; several states in the US have followed suit with their own versions of the rebate, and at least one utility in the UK has run a pilot. The pattern is real, and it is worth understanding before the next request for proposals lands in your inbox.

Why CI Runners Became the New Pipe Inspectors

Leak detection used to be a monthly ritual. A technician would walk a route with a listening stick, or a utility would run a district metered area analysis once a quarter. The latency between a leak starting and someone noticing it could be weeks, which in a pressurized system means thousands of gallons lost and, worse, the quiet undermining of roads and foundations.

Continuous monitoring flips that timeline. With a pressure sensor on a fire hydrant or a smart meter at the service line, data can arrive every minute. But data alone is not insight; you need something to look at it. This is where the CI runner fits. It executes a job that fetches the latest readings, compares them to a baseline model, and flags deviations that exceed a threshold.

The economics have made this possible. Compute that used to cost a few dollars per run has fallen to fractions of a cent, especially when scheduled in off-peak windows. An open-source stack, a PostgreSQL database, and a Grafana dashboard can replace a proprietary SCADA system that required a six-figure license and a dedicated server room. The same YAML that builds your app can now schedule a leak check at 3 a.m. when water demand is lowest.

There is a cultural shift underneath the technical one. Utilities are hiring software engineers, not just civil engineers, and they are discovering that the practices of modern software delivery, code review, automated testing, and deployment pipelines, map cleanly onto infrastructure monitoring. The CI runner is the bridge between two worlds that rarely talked to each other.

The Economics of a Water Meter That Runs on Credits

The unit economics are surprisingly favorable. A single CI run that processes a day's worth of flow data from a dozen sensors might consume a few hundred megabytes of memory and a minute of CPU time. At typical cloud pricing, that is well under a cent. Even a city with a thousand sensors might only need a few hundred runs a day, and the total compute bill can stay below a few hundred dollars a month.

That is a rounding error compared to the cost of a single main break. The American Water Works Association has estimated that non-revenue water, water that is produced but not billed, accounts for an average of 16% of production in the US. A single leak that goes undetected for a month can waste hundreds of thousands of gallons, and the repair cost, especially if the leak undermines a road, can run into the tens of thousands.

The rebate structure nudges this calculation further. By covering the compute credits, the municipality effectively subsidizes the software that makes early detection possible. Hardware is amortized over years; software is a recurring line item. The vendors who sell the sensors now bid on the rebate itself, offering a lower hardware price in exchange for a longer software contract, a model that resembles the classic razor-and-blades strategy.

For the city, the shift from capex to opex is deliberate. Budget cycles are shorter, and usage-based pricing aligns costs with actual consumption. If a district is dry, you are not paying for idle hardware. If a district is leaking, the software is earning its keep. The rebate simply formalizes what the city was going to pay anyway, but it now comes with a line item that a software vendor can point to.

Who Owns the Pipeline? Market Structure and Lock-In

Whenever software enters a historically hardware-dominated market, the question of who owns the data format becomes existential. In the water utility world, the legacy players are the SCADA integrators who have been installing proprietary systems for decades. Their business model relies on long maintenance contracts and closed protocols that make switching costs prohibitive.

The new entrants are small startups, often founded by engineers who came from DevOps or cloud-native backgrounds. They write open-source collectors that speak Modbus or MQTT, and they store data in standard formats like Parquet or InfluxDB line protocol. Their pitch to a utility is simple: you own your data, and if you do not like us, you can take it elsewhere.

That pitch is compelling, but it collides with procurement reality. Utilities are risk-averse, and a multi-year contract with a startup is a gamble. The rebate programs tilt the field by making the software line item visible and reimbursable, which lowers the perceived risk. But the real battleground is the API. Whoever documents a stable, well-supported API becomes the default choice for the next integration, and that position is hard to dislodge.

Early movers are already setting de facto standards. One startup, based in Austin, has published an open specification for leak-event payloads that several utilities have adopted. Another, in the Netherlands, has built a dashboard that aggregates data from multiple vendors, effectively becoming a meta-layer. The market is still fluid, but the pattern is familiar: the first protocol that gains critical mass becomes the incumbent.

For buyers, the lesson is to read the fine print on data rights. Some vendors will try to claim ownership of the aggregated data, or restrict the ability to export it. The rebate rules often require that data be shared with the municipality, which is a good default. But the contract is where the real power lies, and a utility that does not negotiate for data portability is locking itself in for a decade.

The Career Arc: From Deploy Pipelines to Water Pipelines

For an engineer who has spent years optimizing build times and wrestling with flaky tests, the move to civic infrastructure is surprisingly natural. The core skills, writing reliable, observable code, automating routine operations, and troubleshooting under pressure, transfer directly. The domain is different, but the discipline is the same.

The appeal is stability. Municipal contracts run for years, and the work is not subject to the whims of a venture capital cycle. A DevOps engineer who joins a water utility or a civic-tech startup can expect a slower pace, but also a sense of permanence. The projects are not going to get canceled because a growth metric dipped.

Security expertise is a particular differentiator. As water systems become connected, they become targets. The same skills that harden a CI pipeline, secrets management, least-privilege access, and audit logging, are now being written into bid evaluations. A utility that has been burned by a ransomware attack is more likely to pay for an engineer who can demonstrate a threat model.

The boring work of monitoring, the scheduled jobs, the alert thresholds, the log rotation, is suddenly high-value. It is the same unglamorous work that keeps a modern SaaS running, but now it keeps a city's water flowing. Engineers who are tired of fighting over microservices boundaries might find that a water main is a more satisfying adversary.

What You Give Up: The Trade-Offs of Civic Engineering

The trade-offs are real. Release cycles in municipal technology are measured in quarters, not days. A change that would take an afternoon at a startup might require a change advisory board, a security review, and a week of regression testing. The agility that developers take for granted is often absent.

Procurement timelines are another shock. A pilot project can take six months to approve, and a full rollout can take two years. The rebate program helps, but it does not eliminate the paperwork. Engineers who are used to shipping code to production on a Friday might find the pace frustrating.

Salary ceilings are often lower than at a venture-backed scale-up, especially in public agencies. The total compensation might be 10 to 20 percent less, and stock options are replaced by a pension. For some, that is a fair trade for the stability and the mission. For others, it is a dealbreaker.

But the impact is tangible. When a leak is caught early because a CI job flagged an anomaly, the savings are not abstract. They are measured in gallons saved and roads that do not need to be torn up. That is a satisfaction that is hard to replicate in a consumer app. The counter-argument is that the same skills could be applied to a health-tech startup with a higher ceiling. Reasonable people will disagree.

Real-World Deployments and Lessons Learned

To understand how these systems behave in practice, consider a few examples that have emerged over the past two years. In a mid-sized city in Colorado, a civic-tech nonprofit partnered with the water department to deploy a network of pressure sensors on fire hydrants. The initial pilot covered a single district with about 500 connections. The CI runner was configured to fetch data every 15 minutes during off-peak hours, and the anomaly detection model was trained on six months of historical flow data. Within the first month, the system flagged a small leak that had been bleeding for weeks. The repair, costing a few thousand dollars, saved an estimated 1.5 million gallons over the following year. The pilot was so successful that the city expanded it to three more districts, and the rebate program covered the compute costs for the entire expansion.

Another example comes from the Netherlands, where a regional water authority ran a pilot with a startup that used acoustic sensors on pipes. The sensors transmitted data over LoRaWAN, and the CI runner aggregated the data into a time-series database. The system detected a leak that traditional methods had missed for months, and the authority estimated that the early detection saved them roughly €50,000 in repair and water-loss costs. The pilot also revealed a critical lesson: the anomaly detection model needed to be retrained seasonally. Water demand patterns shift with the weather, and a model that worked in the summer produced false positives in the winter when demand dropped. The team added a scheduled retraining job to the pipeline, which reduced false positives by 40 percent.

These examples highlight a common theme: the technology works, but the operational details matter. Data quality is a constant challenge. Sensors drift, batteries die, and network connections drop. A CI runner that assumes perfect data will generate alerts that erode trust. The successful deployments all built in data validation steps, such as checking for stale timestamps and flagging sensors that have gone silent. They also designed their dashboards to show data freshness, so that operators could quickly identify when a sensor was not reporting.

Another lesson is the importance of collaboration between engineers and field staff. Engineers who have never walked a water route may not realize that a pressure spike at 8 a.m. is normal because of fire hydrant testing, or that a certain neighborhood has a history of water hammer. Field staff can provide context that improves the model, and engineers can show field staff how to interpret the alerts. The best deployments I have seen scheduled regular meetings between the two groups, and they used the CI runner's logs to review what had triggered each alert. This feedback loop turned a black-box system into a shared tool.

Finally, the economics are not always as rosy as the early pilots suggest. The compute costs are low, but the total cost of ownership includes sensor maintenance, data storage, and the time of the engineers who maintain the pipeline. A utility that deploys a thousand sensors will need a plan for replacing batteries, recalibrating sensors, and managing the database. The rebate may cover the compute, but it rarely covers the labor. Utilities that fail to budget for this ongoing maintenance will find their monitoring system degrading over time, with sensors going offline and alerts being ignored.

Practical Takeaways for Engineers and Buyers

For engineers considering a move into civic infrastructure, the first step is to learn the data formats. Modbus, MQTT, and the various telemetry protocols are not hard, but they are unfamiliar. A weekend project that reads a temperature sensor and publishes it to an MQTT broker is a good start. The frameworks will come later; the data is the foundation.

Build for auditability from day one. A utility will want to know who changed a threshold, when, and why. Version control, code review, and immutable logs are not optional. The same discipline that makes a PCI-compliant pipeline work will make a water monitoring system trustworthy.

Pilot with one district before scaling citywide. The temptation to roll out a new system everywhere at once is strong, but the prudent path is to pick a small, well-understood area and iterate. The rebate program may even require a pilot phase, which is a chance to build a reference implementation.

Read the fine print on rebate eligibility and data rights. Some rebates require that the data be shared with the utility, which is fine. Others may impose restrictions on the software vendor, such as requiring a local support presence. Knowing these constraints before you sign will save you from a surprise later. The market is young, and the rules are still being written. The engineers who show up early, with clean code and a respect for the domain, will be the ones who shape the protocol for the next decade.