Build the agent people trust with their life.

The work

This is a build-and-defend role, not a policy role. You will model threats, write controls, red-team the live agent, and harden the runtime at the same pace we ship product.

• Threat-model prompt injection, tool abuse, untrusted web content, and cross-tenant exfiltration.
• Harden Firecracker isolation, host and VM boundaries, OAuth tokens, API keys, and keyring custody.
• Build adversarial harnesses, anomaly detection, abuse controls, and permanent regression tests.
• Own incident response from detection to same-day fix, then close the whole class of failure.
• Push toward host-root-resistant privacy with encryption, crypto-shred, and honest guarantees.

You may be a fit if

• You ship security code across application, infrastructure, and host layers.
• You understand agentic threats in practice: injection, jailbreaks, tool calls, poisoning, and leakage.
• You are comfortable with Linux, networking, sandboxing, secrets, Python, and TypeScript.
• You think adversarially and prefer a real control today to a perfect slide deck next quarter.

The work

This is a founding QA role. You will test by hand and in code, define the quality bar, and build the harnesses instead of inheriting a process.

• Cover the Telegram bot, host services, agent runtime, admin console, and Mini App end to end.
• Design evals for accuracy, behavior regressions, latency, loops, refusals, and malformed output.
• Use AI agents to generate cases, drive the product, reproduce failures, and expand coverage.
• Turn escaped bugs into permanent Pytest, Vitest, Playwright, smoke, or live-trace checks.
• Keep release gates sharp enough to catch risk and light enough to preserve same-day shipping.

You may be a fit if

• You combine exploratory product testing with strong Python or TypeScript automation.
• You want to evaluate agent behavior, not only deterministic web interfaces.
• You can build a QA practice from zero and know which failures actually deserve a gate.
• You have sharp product judgment around flows that touch real personal data.

The work

This founding role sits across brand, digital marketing, and PR. You will do the work and build the function, not manage a layer of agencies.

• Define and protect arxi's voice, character, visual identity, and line between confidence and overclaiming.
• Own positioning, landing pages, launches, social, content, SEO, paid experiments, and the activation funnel.
• Build relationships with press, creators, and communities around moments worth talking about.
• Turn real product behavior into stories while avoiding generic AI-agent language and visuals.
• Listen to how users describe arxi and feed that language back into product and growth.

You may be a fit if

• You have real range across brand, growth, content, and PR.
• You can reduce a complex product to one true, memorable line.
• You have strong taste, ship your own work, and measure growth without overselling.
• You want to build a marketing function from zero around a consumer or prosumer product.

The work

We mean full-stack literally. You will move across the Telegram bridge, Python services, web surfaces, VM lifecycle, storage, billing, scheduling, and telemetry.

• Ship end-to-end features in TypeScript and Python, from data model to production.
• Build agent tools and OpenAI Codex plumbing while tracing real turns for accuracy and latency.
• Work on per-user Firecracker microVMs running OpenClaw and the services around them.
• Build considered React, Next.js, and Telegram Mini App interfaces.
• Write tests, open the PR, deploy, inspect production evidence, and own the result.

You may be a fit if

• You are strong in both TypeScript and Python and have shipped production systems end to end.
• You care about frontend craft, backend correctness, security, and latency in equal measure.
• You are self-sufficient, learn unfamiliar systems quickly, and automate your own toil.