Secretlab is an international gaming chair brand seating over three million users worldwide, with our key markets in the United States, Europe and Singapore, where we are headquartered.
This is a senior individual contributor role under our Data & AI team. From day one you run three real surfaces — the eval platform every skill ships through, the MCP (Model Context Protocol) serving layer that exposes products into the chat clients people already use (each user's own permissions inherited), and the skills registry. You're the only engineer on them, so the platform is yours immediately. What you grow into is owning the function — calling the shots and carrying the risk. Platform strategy and architecture direction sit with the AI Engineering Lead; implementation decisions on your surfaces are yours. No management ladder to fight. And instead of vague promises: a written 12-month scope review with pre-agreed criteria.
How we size the work — the A–D ladder
We size a build not by how hard the code is, but by how much human work it replaces, and at what reach:
- A — Task augmentation: speeds up one task; the human still drives.
- B — Task replacement: a loop that does one whole complex task end-to-end; the human reviews the output instead of doing the work.
- C — Role replacement: a full agent covering a person's whole workload under governed autonomy (drafts, never merges, until it earns reach).
- D — Company-scale swarm: many agents as one system at org scale — the hard problem is the swarm (communication, work distribution, blast-radius governance across the fleet).
Building a skill is not the signal. The signal is where on the A–D ladder you can build, own, and operate.
To be successful
Four outcomes, in priority order. The first two are the floor, the third is success, the fourth is exceptional.
- Build fast, and build correctly. Ship real AI products at pace. Every skill or assistant passes its eval gate before it ships — no eval, no ship. Slow is a problem; wrong is worse. We need both.
- Ship and operate the bones of the platform — with the AI Engineering Lead. Build on and operate the eval platform, the skills/agent registry, and the MCP serving layer, and the governance defaults every future agent inherits. You own the implementation and the operation. You'll carry the pager here — we don't require that you've carried one before.
- Raise the team's AI-nativeness by showing, not telling. Make the data team more AI-native by building the tools and patterns that upgrade them. Don't run workshops. Level up one senior multiplier who carries it to the wider company.
- Ship the first tier-B/-C loops that take whole tasks off the team's plate. End-to-end loops — a documentation-maintenance pass, a code-review or test-writing loop — aimed at our own delivery velocity first, drafts-never-merges.
What your week looks like
- Build blocks on the platform (assistant, serving layer, eval platform) — protected time; this is the core of the role.
- Eval runs and regression gates on every skill or agent change shipping that week — yours or the team's; you operate the gate.
- Production ops: usage/cost telemetry review, incident follow-ups, silent-failure sweeps — you operate what you ship.
- Team enablement (capped, a few hours/week): build the tooling/patterns that make the data team more AI-native, and pair with one senior multiplier who carries it to the wider company. This is show-and-tool, not train-the-company.
- Product pairing with a business domain expert — encoding their judgement into an agent or agentic loop. One-off skills aren't this role's job: the team builds those on the platform you run. You build the agents and the loops, and you manage the infrastructure they run on.
- Weekly sync with the AI Engineering Lead on architecture direction and use-case prioritisation.
Requirements
We hire a spiky profile on purpose: deep where it matters, honest about the rest. Three tiers — the base (need it), bonus (moves you up the list), and grow into (not expected at hire). What this role is not comes last.
The base — need it
- A real engineering foundation: 4–7 years building software or data systems before the AI pivot. Python, solid API work, comfortable in SQL, and the habits that come from running code other people depend on.
- Has automated real parts of their own work — systems they actually use every day, and can explain why they're built the way they are. If you don't build leverage for yourself, you won't build it for a team.
- Makes the people around them better with AI by building them the tool.
- Can show how they know their system works. Any method that produces evidence counts; confidence doesn't. It seemed fine is a no.
Bonus — moves you up the list
- Shipped and operated LLM or agent systems in production — real users, real incidents, on-call and all. The biggest bonus on the sheet. It is not a gate: if the base is there and this isn't, we still want to talk.
- Formal eval experience — test sets, automated grading, regression checks on model changes. We run this discipline in-house and will teach it; arriving with it means you start faster.
- Snowflake Cortex (Analyst semantic views, Search, Agent Skills, managed MCP server) — trainable; we care how fast you learn it, not whether you already have.
- Multi-agent / swarm orchestration — matters more as we build up the ladder; not needed now.
- Has cut an AI bill before. Useful, not required — we build fast first and handle spend at the platform level.
- Observability/tracing for agents; multi-provider experience; dbt; Slack or Google Workspace APIs.
Grow into — not expected at hire
- Owning the function — not just the platform. The platform surfaces are yours to run from day one; you're the only engineer on them, so that part is immediate. What you grow into is owning the function: calling the shots, taking the risks, falling on the sword when it breaks. That's earned. We are not looking for someone who has already run a function of their own.
- Making the wider organisation AI-native — that reach runs through the senior person you level up, not through you directly.
Personality
Everyone here is held to the same five department expectations: run AI-native, own what ships, solve the business problem, say it straight, leave the process better than you found it. The role adds the technical bar on top.
- Builder who upgrades the people around them — energised by making the team more capable, and does it by building the thing that teaches.
- Upfront and candid — honest about capability without embellishment; comfortable with direct feedback.
- Pragmatic shipper — MVPs, avoids over/under-engineering, knows when NOT to build an agent, and doesn't thrash the design.
- Content as a senior IC with owned surfaces — we're upfront about the shape of the role from the first call.
Explicitly not this role
- Research scientist / PhD / publications track — the mandate is composition on foundation-model APIs, not training.
- Model fine-tuning specialists.
- Prompt-tinkerers without an engineering base — can talk agents, can't build them.
