Agent Platform Engineer

The Short Version

You make everything the agent runs on solid, and build the systems that keep it improving on its own: the integrations it reaches through, the gateway that turns them into functions it can call, the sandboxes it runs in, and the data underneath. You've built and operated real infrastructure at scale and have scars from how it breaks. You may never have shipped an LLM feature, and that's fine. If you haven't run systems in production, this isn't the role.

The Hard Part

Viktor reaches into 3,000+ integrations for 25,000+ workspaces, 1.5M+ tool calls a day, on real customer data. Every integration that quietly misbehaves is a task the agent gets wrong.

We're betting one general agent across everything a company runs on beats a stack of narrow tools, and that bet lives or dies on the infra underneath. Thousands of integrations can't be hand-maintained, so your real job is the machine that makes them improve on their own, on top of the gateway, sandboxes, and data they all run through.

What You'll Actually Do

• Build the self-improvement machine for integrations: agents that explore each API and learn it, skill files that self-correct when a tool breaks, and a loop that turns every failure into a permanent fix.

• Own integration quality end to end: catch issues before customers do, fix them, and make sure they can't recur.

• Build the tool gateway: turn every connected API and MCP into a clean Python SDK the agent composes, with routing, auth, rate limits, and reliability.

• Run the sandboxes: a persistent Linux environment per workspace, isolated, secured, autoscaled, and cheap per task.

• Own the data layer: state and persistence across 25,000+ multi-tenant workspaces, durable and fast.

• Whatever needs building. Small team, large surface.

The Bar

You ship to production every day, and the changelog has your name on it. When a sandbox leaks or an integration starts failing at 3am, you fix it, because you understand the system end to end. You design for how things break at scale, with blast radius, failure modes, and cost in your head before you ship.

Who You Are

• You design distributed systems for scale and reliability, and have operated them in production through the breakage.

• You know sandboxing and OS-level isolation cold: containers, Linux internals, running untrusted code safely.

• You've wrangled real integration plumbing: OAuth, webhooks, rate limits, and third-party APIs that lie.

• You think in multi-tenancy, durability, and cost per request, and design for failure by default.

• Security and blast radius are instincts you bring to every design, especially when the thing executes model-written code.

• You build with AI by default, because that's how the team moves.

• You may never have shipped an LLM feature, and that's fine. This is a systems role.

Why This Role Is Different

• No layers. You work directly with both founders, and decisions get made in the room, not in a Linear ticket.

• The platform is the ceiling. Every integration you make reliable and every sandbox you make cheaper raises what the whole agent can do.

• Scale forces the bar up. 1.5M+ tool calls a day means a shortcut breaks in production the same week.

Even Better If

• You've worked on infrastructure, platform, or developer-tools teams where reliability was the product.

• You've built sandboxing, code-execution, or multi-tenant systems before.

• You're at home in AI-assisted, spec-to-production engineering workflows.

• You've contributed to the open-source infrastructure the world runs on.

• You've founded something or been an early-stage builder.

Tech

Python on the backend and agents, Modal for infrastructure, persistent Linux sandboxes per workspace. You'll live in container orchestration, OAuth and webhooks, the tool gateway, and multi-tenant data. You don't need all of it coming in, but you need to learn fast.

How we work

Small team, high trust, low process. Decisions are made by owners, not committees. You will ship your first week. You will talk to users your first day.

We don't do alignment meetings or stakeholder syncs. We build things, see if they work, and iterate.

Why Viktor

We're one of the fastest-growing companies in the world. The product works. The market is pulling.

This is a rare window: everyone here owns something real. Not a task. A surface of the company that customers depend on.

That doesn't last forever. Right now, it's still true.

Compensation

Top-of-the-market salary and the kind of ownership that only exists at this stage.

The best work happens when you're in the room. Munich, New York, Dublin, and Warsaw. Remote for some roles.