Software Engineer

Nyotta AI Inc.; San Francisco (CA); On-site; USD 140,000 - 230,000; Full time

Job summary

Nyotta AI Inc. is hiring in San Francisco to design runtimes for AI agents, orchestrate tool usage, and manage multi-step workflows end-to-end. You will build scalable runtimes, interfaces, and integrations that ensure reliable execution across model calls and external services.

The role emphasizes clarity, debuggability, and user-facing review interfaces. You will collaborate with ML engineers to improve latency, cost, and reliability while shaping the product experience for directing and

Qualifications

• Experience building software with clear design decisions.
• Proficient in Python or TypeScript and working with APIs and data models.
• Able to explain tradeoffs and failure modes in complex systems.

Responsibilities

• Design agent runtimes that coordinate model calls, tools, task state, and multi-step execution.
• Build APIs and integrations that give agents access to data and actions within explicit permission boundaries.
• Handle retries, interruptions, and partial failures without losing progress or repeating actions.
• Build traces and monitoring to understand agent activity and where execution went wrong.
• Create interfaces for directing agent work, reviewing results, and approving consequential actions.
• Improve runtime latency and cost, validating changes with ML engineers before release.

Skills

Software design

Debugging across components

API usage

Tools

Python

TypeScript

Job description

Design the systems that let agents plan, call tools, and carry work through to completion. Build the runtime, integrations, and interfaces that make those workflows dependable.

Nyotta builds AI agents for complex engineering and business workflows. Our work brings together reasoning over technical information, tool use, and human review so teams can move from source material to decisions and completed work. We apply these systems to engineering and commercial processes where accuracy, traceability, and control matter.

About the role

You will build the software around the models: execution state, tool interfaces, orchestration, and the product experience people use to direct and review agent work. A task may involve multiple model calls, external services, and a human decision before it can proceed. You will work with ML engineers to make that execution understandable, recoverable, and straightforward to improve.

In this role, you will

• Design agent runtimes that coordinate model calls, tools, task state, and multi-step execution.
• Build APIs and integrations that give agents access to data and actions within explicit permission boundaries.
• Handle retries, interruptions, and partial failures without losing progress or repeating completed actions.
• Build traces and monitoring that help the team understand what an agent did and where execution went wrong.
• Create interfaces for directing agent work, reviewing results, and approving consequential actions.
• Improve runtime latency and cost, and work with ML engineers to validate changes before release.

Your background

• You have built software you can explain in detail: its design, tradeoffs, and failure modes. Personal projects, internships, and professional work all count.
• You write Python, TypeScript, or another backend language and are comfortable working with APIs, services, and data models.
• You can debug a problem across multiple components and make unfamiliar systems easier to understand.
• You care about how people use the software and can translate an unclear workflow into a practical implementation.

Level and scope

• Engineer: Build a scoped runtime, integration, or product feature with guidance and regular code review. Early-career applicants are welcome.
• Senior: Own a system from design through release and operation. Bring experience making architecture decisions, handling production failures, and improving reliability over time.

Helpful experience

Experience with agent frameworks, workflow engines, distributed systems, asynchronous jobs, observability, or enterprise integrations.

What makes this work hard

Agent execution is unpredictable: a model may choose an unexpected path, a tool may time out, or a task may pause for human input. The system needs to preserve state and make those cases visible without duplicating actions.

A problem you might work on

A tool times out after accepting an action. Design how the agent checks what happened, resumes safely, and gives the user a clear account of the outcome.