Member of Technical Staff, Agent Infrastructure and Reliability

What we're building

Enterprise marketing still runs through agencies because launching one campaign means coordinating specialists across Marketo, HubSpot, Meta, LinkedIn, CMSs, and a dozen other systems. It takes weeks.

CharacterQuilt uses computer-use agents to do that work inside the software companies already use. A campaign that took three months can now ship in two days. We raised $13M and were one of the fastest-growing companies in YC's P26 batch.

The role

You'll own the runtime behind every agent run: workers, sandboxes, files, skills, authenticated browsers, durable state, retries, recovery, and the API and CLI around the system.

You'll work directly with me and a very small engineering team. Our agents touch customer software and real budgets, so "probably worked" is not good enough.

What you'll do

• Provision and operate workers and sandboxes.
• Make file, skill, and workspace hydration fast and correct.
• Build reliable browser authentication and session isolation.
• Design retries, replay, cancellation, recovery, and observability.
• Ship the API, CLI, and MCP surfaces customers and our own agents use.

You might be a fit if

• You're a strong full-stack engineer who likes databases, distributed systems, filesystems, and browsers.
• You move quickly but care deeply about correctness.
• You want broad ownership at an early-stage company.

Five days a week in Union Square. $250K to $350K base plus 0.5% to 3% equity. We do not sponsor US visas.

Interview Process

1. A founder conversation about your best work, what you want next, and the CharacterQuilt mission.

2. A focused technical screen using one real agent-lifecycle failure.

3. A systems deep dive on durable execution, browser state, publication, and recovery.

4. A paid two-day onsite work trial on a bounded, production-shaped problem.

5. References and a direct offer conversation.

We will share the purpose and evaluation criteria before every stage. The work trial is paid and is designed for mutual evaluation. You should leave with a real view of the code, the pace, and what working with the founder feels like.