Software Engineer

• Own the agent harness at the core of the product: the workflows, checkpoints, and evaluation that keep long-horizon agents reliable across hours of testing
• Design and own the backend services and APIs that carry evidence from ingestion, through analysis, to auditor-ready output
• Build integrations with the systems our customers keep evidence in: cloud infrastructure, identity providers, ticketing systems, ERPs, and more
• Design the data models and system architecture that have to hold up as our customer base and control volume grow
• Write code and design systems with security and data integrity in mind, given the sensitivity of the evidence our customers share with us
• Partner directly with our customers to craft and ship the features that matter most to the people actually doing the testing

• Experience building on LLMs and agentic systems: harnesses, long-horizon agents, evals, or resilient workflows that survive contact with messy real-world input
• Experience building enterprise SaaS products that handle sensitive customer data
• Comfort working across the stack: solid backend fundamentals (APIs, data modeling, distributed systems) plus enough frontend fluency to ship a feature end to end
• Experience with AWS, Kubernetes, and infrastructure as code
• Experience in fintech, audit, governance, risk, compliance, or another regulated environment
• A track record of high ownership at an early- or growth-stage startup, where you shaped not just the code but the product and the process around it

Petual is the AI-powered control tester for the modern enterprise. Our agents test whether a company’s financial, operational, and technology controls actually work, and trace every conclusion to its source.

The hard part isn’t speed. It’s being right. External auditors re-perform our work, so precision on sample 1 has to equal precision on sample 400, and an agent that can’t support a conclusion has to say so rather than guess. We’re backed by Andreessen Horowitz; the team comes from Stripe, Retool, and Lyft; and public companies already run their SOX programs on Petual.

The Opportunity

The interesting problem is not getting an agent to test a control once. It is knowing whether it was right across hundreds of controls, for customers whose methodologies all differ, where the ground truth is a senior auditor’s judgment rather than a label in a dataset. Building the harness and the evaluation that make that tractable is the core of this role, and it would be yours to own.