How Voxel engineers keep safety systems in check with Firetiger

Jorge Villatoro is a principal engineer at Voxel, an industrial workspace safety and analytics company.

Jorge and I have known each other for over a decade, he's one of the engineers I trust most when it comes to infrastructure and reliability. So I was delighted to learn about the impact that Firetiger had on how his team handles production issues.

Voxel takes camera feeds from industrial sites like warehouses, factories, or construction yards, analyzes them in realtime for safety compliance. Risky Forklifts operation, vehicles or workers navigating the workspace. They're using AI to watch for risky patterns and report metrics to site safety managers to make the operations safer over time.

When you're on the critical path of physical safety operations, the bar on reliability is extremely high.

When the measurement system is the product

Most reliability conversations are about uptime. Is the service responding? Are requests succeeding?

But Voxel's reliability challenges are more subtle than they look, and Jorge explained it very well:

"At the end of the day, our customers expect that the numbers they get from us are meaningful, and that the changes in those numbers are meaningful.

Anytime our reliability changes, it can have a big impact on the statistics we give a customer. They need to be able to trust that when the numbers on their dashboard change, it's because something actually changed at their site."

That's a different kind of stakes than "the API is down." When Voxel's systems are under stress, the output contains noise that makes the data untrustworthy.

Reliability, for Voxel, is all about preserving meaning of the data.

Making sausage, not sausage machines

Voxel's engineering team is deep in computer vision and data analytics; that's the core of their product and where they invest to deliver value to their customers.

Operating the kind of distributed systems that power their product takes a whole different type of skills, and Jorge has a great way of putting it:

"We want to hire people making the sausage, not making sausage machines.
[...]
Firetiger helps level up engineers who are not as proficient in operations and reliability, and let them get from 'something appears to be wrong' to 'here's a potential root cause,' and sometimes even allow them able to resolve issues autonomously."

When something breaks in production, the evidence is there. Logs, metrics, anomalies. But the work of collecting that evidence and tracing it back to a root cause is a specialized skill.

Most of his engineers are versed in Voxel's core expertise. With Firetiger's assistance, they are empowered to own a larger scope of the engineering realm.

Onboard it like an engineer

Jorge treated Firetiger like a new teammate: walked it through the codebase, the metrics, the tools, the same way you'd onboard a senior hire. His framing is exactly right:

"I guided it through the repo, the different metrics and tools... just like onboarding an engineer, honestly!"

That investment pays off every day. Often times, Voxel's alarms are noisier than they'd like, and engineers were spending real time figuring out whether a page meant anything.

"We fire off a one-shot query, Firetiger pulls logs and gives a clear assessment.
[...]
When something actually is broken, it hits the root cause pretty quickly, usually within a couple of prompts."

The agent operates like capable engineer who knows your system really well.
With the right context, it can feel like magic!

Where we're heading

Jorge sees two groups using AI tools, with various level of results:

1. Experienced engineers who know how to evaluate the output critically.
2. Less experienced folks shipping 2000-line PRs they can't really review, handing them to engineers who can.

Jorge's prediction is unsentimental:

"Whether anybody likes it or not, more code that is written or triggered by people who don't really understand the code is going to end up landing.

And that means reliability is going to become more of a challenge over time. Having tools that help you manage that risk is pretty critical."

That's the future Firetiger is built for.

More AI-written code in production, and a reliability surface that keeps expanding. Not a gatekeeper, but as a continuous safety net: reactive when something breaks, and increasingly proactive as the system builds deeper context about what normal looks like.

For Voxel, the stakes are higher than most, but it's a clear case of how pairing engineers with agents clearly augments their abilities to meet those challenges.