Staff Software Engineer, AI Agent Platform

Light & Wonder builds games, cabinets, and platforms behind a large share of the world’s regulated gaming floors and online casinos. We are standing up a small engineering team in Austin to build production agentic systems to address some of the least glamorous yet most valuable problems in our business.

If you have seen the forward-deployed engineering model at Palantir or at the AI labs, this is that pattern pointed inward. You’ll be embedded with a business function, learn how the work gets done, and ship a real system against it. Then you do it again somewhere else, and the platform gets richer each time.

Two of the first initiatives on our roadmap:

• Jurisdictional licensing. Working in gaming means our people hold occupational licenses, frequently across many regulators at once. Every jurisdiction wants overlapping but subtly different slices of the same personal and employment history, in its own format, with real consequences for getting it wrong. There is no canonical source of truth to draw from, so this is done by hand repeatedly today. We are building agents that gather, reconcile, and pre-fill these applications, with the employee reviewing and approving before anything is submitted. It is not glamorous. It is a hard reconciliation problem wrapped around a real human cost, and fixing it gives time back to colleagues who will notice.
• Multimodal part discovery. A field technician standing in front of a cabinet should be able to describe or photograph a component and get the right part number, compatibility notes, and availability. Today that lookup depends on tribal knowledge and a fragmented catalog.

Neither of these is a demo. Both have real users, real deadlines, and, for anything touching the regulated side of the business, real expectations about how an automated decision can be explained after the fact. What is hard about this

We would rather tell you now than in week three.

• Data quality. The source data is inconsistent, incomplete, and spread across systems that were never designed to be read by anything but a person.
• Fragmented systems. There is no single API to point an agent at. A meaningful share of the work is building consumable surfaces where none exist.
• Tool sprawl. Agents with fifty tools do not work. Deciding what an agent should be allowed to touch, and when, is a design problem we have not solved yet.
• Human-in-the-loop approvals. Most of these workflows cannot and should not be fully automated. Designing approval handoffs that people trust and use is as much of the job as the orchestration.

The team

Today the team is our VP of Engineering, who is currently the principal developer on this codebase, and a technical program manager. A DevOps/MLOps engineer joins shortly. We are hiring three engineers for this team at both Senior and Staff level, and we will calibrate the level based on what you bring.

That means a short path from decision to shipped work, with nobody translating between you and the person setting priorities. Some engineers find that the best environment they have worked in; others want more people around them. We would rather you know which you are before you apply.

The codebase is early enough that you should treat it as greenfield. The decisions about platform primitives, eval standards, and orchestration patterns have not been made yet; whoever joins now will make them.

We operate like a startup inside the enterprise: shipping over process, evaluation over opinion, and platform investment that makes each deployment faster than the last. We are anchored on a modern Microsoft stack, built around Microsoft Agent Framework (.NET), Azure, and a data backend spanning Snowflake, Databricks, and Microsoft Fabric. About you

You own agentic workflows from the interface down to model execution: React on the frontend, a C#/.NET orchestration layer behind it, and whatever integration work the tools demand. You are responsible for the orchestration logic, the multi-agent handoffs, and the eval coverage that makes behavior predictable enough to ship.

We do not expect experience with Microsoft Agent Framework specifically. It reached 1.0 in April 2026 and almost nobody has real depth in it. We do expect that you can learn an unfamiliar framework quickly and write production-grade code while doing it, and we will test that directly in the technical interview loop.

One thing worth being honest about: this is non-deterministic software. If you are most comfortable when correctness is binary and testable, this work will frustrate you. The engineers who will do well here can look at a workflow that succeeds 94% of the time, characterize the other 6%, and make a defensible call about whether it ships. What you will do

• Architect and deliver agent workflows end-to-end, spanning conversational UI, orchestration logic, and tool integrations.
• Build embedded agent interfaces that feel like products rather than chat boxes.
• Implement orchestration in Microsoft Agent Framework (.NET): tool calling, multi-agent patterns, and human-in-the-loop handoffs.
• Develop behavioral, regression, and adversarial evaluations for everything you ship.
• Contribute the shared primitives, reusable templates, and review standards the rest of the team builds on.

• 5+ years of professional software engineering experience, with real production depth on both the front and back end.
• Strong C# and .NET, including modern .NET (8 / 9 / 10), async patterns, and dependency injection. This is the primary stack, and it is required, not preferred.
• Strong TypeScript and React.
• At least one production system using LLMs, shipped end-to-end. “Production” means real users, real reliability constraints, and real evals, not a hackathon project or a tutorial follow-along. Multi-step orchestration counts for more than single-completion-call work.
• Experience writing evaluations for non-deterministic systems: behavioral tests, regression harnesses, human-rated samples., * Azure primitives (App Service / AKS / Functions, Key Vault, API Management, Entra ID) and comfort writing infrastructure as code.
• Agent orchestration in any ecosystem: Microsoft Agent Framework, Semantic Kernel, AutoGen, LangGraph, or the OpenAI Agents SDK.
• Agent or copilot UI frameworks such as AG-UI, CopilotKit, or comparable.
• Time on a platform or developer-experience team. You know what reusable looks like., Bachelor’s degree in computer science or a related field, or equivalent professional experience.

• Hybrid schedule anchored in our Austin, TX office with 4 days per week in the office. Occasional travel to L&W sites and stakeholder locations.
• Relocation support is available for candidates outside Central Texas.
• Compensation: Your recruiter will share the range for this role on our first conversation.

Light & Wonder and its affiliates (collectively, L&W) are engaged in highly regulated gaming businesses.As a result, certain L&W employees may, among other things, be required to obtain a gaming or other license(s), undergo background investigations or security checks, or meet certain standards dictated by law, regulation or contracts. In order to ensure L&W complies with its regulatory and contractual commitments, as a condition to hiring and continuing to employ its employees, L&W requires all of its employees to meet those requirements that are necessary to fulfill their individual roles. As a prerequisite to employment with L&W (to the extent permitted by law), you shall be asked to consent to L&W conducting a due diligence/background investigation on you.

This job description should not be interpreted as all-inclusive; it is intended to identify major responsibilities and requirements of the job. The employee in this position may be requested to perform other job-related tasks and responsibilities than those stated above.