- Pay
- $245–318k
- Where
- Mountain View, CA, USA
- Level
- Senior
The opportunity
Three billion people a month play something built in Unity. We're the team changing how those things get made — building AI that generates assets, writes and repairs code, and turns an idea into something playable faster than has ever been possible.
Our group owns the backend and operations behind all of it: the cloud infrastructure our AI products run on, the data and experimentation platforms that tell us what works, the evaluation systems that tell us whether the AI is actually any good, and the operations that keep everything running across regions and customers.
We're looking for a Staff Software Engineer who is genuinely AI-native — for whom agents and AI tooling are how the work gets done rather than a productivity add-on — and who sets direction rather than waiting for it. We've built the foundations over the past few months — CI/CD, infrastructure-as-code, deployment standards — and they work. The platform is stable, and we intend to keep it that way. What we have not solved is how much of understanding it still depends on people: when a metric moves, someone works out what it means; when something degrades, a human reads dashboards and makes the call; when a model changes, we find out more slowly than we should.
None of that scales with the speed we're now shipping at, and none of it needs to stay manual.
What makes this a Staff role rather than a senior one is that several of these problems do not yet have a defined shape, and we are asking you to give them one. We have more open directions than we can pursue. Deciding which ones are worth the team's time, convincing people you are right, and then building the thing — that is the job.
What you'll be doing
Own how this team builds software with AI. We use AI across our workflow today and we're not satisfied with how far it goes. You'll set the standard — agentic code review and repair, generated tests that stay honest as requirements change, the tooling and patterns the team adopts because they make the work faster. Not just working this way yourself, but moving the floor for everyone, and having the judgment to tell a tool that earns its place from one that doesn't.
Define what we measure, and rebuild observability so it explains itself. Our telemetry has gaps, and much of what we collect is ambiguous, unlabelled, or only meaningful to whoever built it. You'll own the model of system health and business outcomes across our platform, make each metric self-explanatory, and replace manual review and manual reaction with automated interpretation.
Make diagnosis agentic, and decide how far that should go. Build the troubleshooting layer that correlates signals, proposes root causes, and takes safe automated action — so that understanding a production issue doesn't start with a person reading graphs. Where the automation stops and a human stays in the loop is a judgment call with real blast radius, and it's yours to make.
Own the infrastructure side of evaluation, as the technical counterpart to our Quality Engineering Lead. Our products generate code, find and fix bugs, and produce art and design — output with no single correct answer. We have the beginnings of an evaluation system and it isn't mature yet. You'll make evaluation fast, repeatable and cheap enough to run continuously, working as the engineering peer of the lead who owns the methodology.
Decide what AI does for gaming LiveOps. Live games are managed continuously after launch: events and content, player behaviour and retention analysis, monetization, live health. We think Unity's data puts us in an unusual position to make that faster, smarter and far less manual. We are early here and genuinely do not know yet what it becomes. We're not asking you to execute a plan — we're asking you to form one, defend it, and know when to abandon it.
Make agentic systems affordable. As more of our product and our own workflow runs on inference, the economics become an engineering problem — token…