We hire frontier engineers.

You’ll build what people actually touch — Spaces, the editor, the extensions they reach for daily — and own each one from rough idea to the thing in someone’s hands.

What you’ll own

• Design and ship product extensions and core UI, prototype to production, often solo.
• Turn fuzzy problems into shipped features fast, then sharpen them with real use.
• Sweat interaction, latency, and the feel of a tool people use all day.
• Use Frontier to build Frontier, and feed what’s missing back into the platform.

You might be a good fit if

• You’ve owned real features end to end, and people liked using them.
• You’re fluent in TypeScript and React (or pick up a stack in a weekend).
• You hold craft and pace at once, and know when to trade one for the other.
• You run several things in parallel without losing the thread.

Bonus

• Developer tools, IDEs, or extension systems.
• Taking a product 0→1 at a small company.

Apply → Copied [email protected] ✓

You’ll own the host: the scheduler, the session and workspace model, the daemon on every machine, and the typed bus everything rides. The part that has to just hold.

What you’ll own

• Scheduling and dispatch: many sessions across a machine pool, queued and routed without the user thinking about it.
• The worker daemon and the machine / workspace model — connection, health, isolation.
• The message bus and persistence every extension depends on.
• Reliability, latency, and the failure modes of running agents across a fleet.

You might be a good fit if

• You’ve built systems other people relied on, and you think in failure modes and backpressure.
• You like infrastructure that disappears, the part people stop thinking about because it holds.
• You’re comfortable across Node/TS, sockets, processes, and containers.
• You go deep on systems while keeping many fronts moving.

Bonus

• Scheduling/orchestration, real-time systems, or multi-tenant isolation.

Apply → Copied [email protected] ✓

You’ll own the brain — how Frontier drives LLMs into useful, steerable, multi-step work: the agent loop, the prompting, the evals, and the “thickening” that makes a turn worth more.

What you’ll own

• The agent loop: planning, tool use over MCP, spawning subagents, returning at the right moment.
• Information-boundary work — more done per turn, fewer interruptions.
• Evals and guardrails: knowing an agent is good, and catching when it drifts.
• Provider-agnostic model integration (Claude, GPT, Gemini, local), getting the best from each.

You might be a good fit if

• You’ve built LLM systems that worked in the real world, not just demos.
• You’re equal parts researcher and engineer — you prototype, measure, and ship.
• You have strong intuitions for prompting, tool use, and agent design.
• You read the field weekly and fold the best ideas in fast.

Bonus

• Eval design, fine-tuning/RL, multi-agent systems, or speech (STT/TTS).

Apply → Copied [email protected] ✓

Frontier runs AI agents that execute real code on real machines. You’ll own making that safe — isolation, the trust model, and the failure cases nobody else is thinking about yet.

What you’ll own

• The isolation and sandbox model for agent-generated code and workspace boundaries.
• The trust and permission model across machines, providers, and extensions.
• Threat modeling for agentic systems: prompt injection, tool abuse, supply chain, exfiltration.
• Secure defaults baked into the platform and the extension API, not bolted on.

You might be a good fit if

• You’ve done security engineering on real systems, not just audits.
• You think like an attacker and build like an engineer.
• You’re comfortable with the weird new threat surface of AI agents.
• You move at startup pace without lowering the bar.

Bonus

• Sandboxing/containers, AppSec, or AI/agent security research.

Apply → Copied [email protected] ✓