Agent Maestro

Superconductor; California (MO); Hybrid; USD 140,000 - 210,000; Full time

Benefits offered by this job

Flexible PTO

Parental leave

Health insurance

Employee-friendly equity terms

Premium equipment

Unlimited token budget

Job summary

Superconductor is building a small, high-standards AI software studio where agents pursue work autonomously across emails, bug reports, sessions, and more. We are a tiny team with massive impact and a practical, token-budgeting approach to tooling.

The Role focuses on advancing multi-agent systems, experimental frameworks, and scalable architectures. You will own significant acceleration projects, iterate fast, and ship features that push our platform forward in ways users rely on daily.

Qualifications

• 3+ years of professional engineering experience, with at least 1 year deeply focused on AI/LLM applications.
• Experience building multi-agent systems or complex LLM-powered applications in production.
• Strong systems thinker who can reason about failure modes, edge cases, and graceful degradation.
• Self-directed and experimental — you don't wait for a roadmap, you build prototypes and demonstrate what's possible.

Responsibilities

• Build intelligent orchestration that assigns the right type of agent (PM, coder, QA, deployer) to the right task.
• Develop self-healing workflows that detect when agents are blocked, diagnose the issue, and either resolve it automatically or escape to humans.
• Experiment relentlessly with new models, agent frameworks, prompting strategies, and tool-use patterns.
• Build evaluation and benchmarking systems, and make numbers go up on them.
• Produce high quality, clean, human-reviewable code — and review a lot more code than you write.

Skills

Python

TypeScript

Ruby

Rails

Tools

Rails

Job description

About Superconductor

Superconductor is building the autonomous software organization — a platform where agents don't wait for instructions, but find work everywhere: in customer emails, bug reports, session replays, Slack threads, X posts, etc. Agents that work while you sleep.

We're a team of 7, backed by $7M in funding, and we intend to stay below ~25 humans hopefully forever. That means every hire matters enormously, and we compensate accordingly with highly competitive salary and equity.

We have been building software together for over a decade. We formerly co-founded Gradescope, an AI-powered education grading platform used by millions of students worldwide. Previously, we were AI PhD students at UC Berkeley. All in all, we are four developers, one designer, and two generalist operators, and we all wear multiple hats every day.

Our Values

• Customer obsession. We see every customer interaction — every support email, every bug report, every piece of feedback — as a chance to earn trust and build something better. Customer communication is our differentiator, not overhead.

• Speed with stability. We move fast, but we do not break things. We know that if something breaks, we're the ones fixing it — sometimes at night. We'd rather invest in getting it right the first time.

• Living in the future. We use Superconductor to build Superconductor. We adopt our own tools before anyone else and push them until they break. If something feels like the old way of building software, we question it.

• Small team, high standards. Every person on the team has outsized impact, outsized ownership, and outsized responsibility.

Working at Superconductor

• Hybrid from our East Bay office: 3 days in our easily BART-accessible office, 2 days from home.

• However, our agents work 24/7, and there's real value in having humans cover the night shift.

• Compensation: Highly competitive salary and meaningful equity — we're a small team that plans to stay small.

• Benefits: Flexible PTO, parental leave, health insurance, employee-friendly equity terms, and premium equipment.

• And of course, a practically unlimited token budget!

The Role

You live and breathe AI agents. You've tried every coding agent harness, and you have strong opinions about what works and what doesn't. You spend your weekends experimenting with agent orchestration frameworks, MCP servers, and multi-agent workflows.

We have less of a clear idea for what this role entails than for our other open roles. Things move so fast that the main attribute we’re looking for is raw intelligence, boundless energy, and an extreme eagerness to learn something new every day.

In your first month, you might take Superconductor to #1 on Terminal-Bench. Or you might use our platform to rebuild a major piece of software in days. Or you’ll develop and evaluate a novel agent orchestration framework. Or you might build something else we haven't imagined yet — pitch us on it!

What You'll Do

• Build intelligent orchestration that assigns the right type of agent (PM, coder, QA, deployer) to the right task

• Develop self-healing workflows that detect when agents are blocked, diagnose the issue, and either resolve it automatically or escape to humans

• Experiment relentlessly with new models, agent frameworks, prompting strategies, and tool-use patterns

• Build evaluation and benchmarking systems, and make number go up on them

• Produce high quality, clean, human-reviewable code — and review a lot more code than you write.

You Are

• Experienced building multi-agent systems or complex LLM-powered applications in production

• Comfortable with all of Python, TypeScript, Ruby

• A strong systems thinker who can reason about failure modes, edge cases, and graceful degradation

• Self-directed and experimental — you don't wait for a roadmap, you build prototypes and demonstrate what's possible

• 3+ years of professional engineering experience, with at least 1 year deeply focused on AI/LLM applications

Bonus Points

• You've built or contributed to an agent framework, MCP server, or agent evaluation benchmark

• You've shipped a product where AI agents operate autonomously

• Experience with Rails is a plus — our platform is built with it