• Platform services: design, build and run the backend services that receive events, track cases, decide whether an agent should run, and resume work when a person responds.

• Correctness: own the platform's correctness guarantees:

  • idempotency;

  • deduplication of repeated signals;

  • locking against concurrent runs;

  • recovery from crashed or stuck runs;

  • a complete audit trail of what each agent did.

  • Agents: build and evolve agents that follow business SOPs. Decide which steps belong in deterministic code, which need the model's judgement, and which must stay with a person.

  • Tools: build MCP tool servers that give agents safe, well-described access to internal data and services, with clear input and output schemas.

  • Human-in-the-loop: integrate approval flows: request approvals, verify signed callbacks, handle partial approval, rejection and expiry, and make sure only approved work is executed.

  • Evaluations: write evals for agents and tools (golden datasets, regression suites, trace review) and use them as the gate for shipping prompt, model and tool changes.

  • Observability and cost: instrument services and agents with tracing, metrics and logs. Track model cost per case, and keep agents from running when nothing has changed.

  • Safety: apply guardrails, including:

    • treating external content as untrusted input;

    • keeping money and policy rules in code, not in prompts;

    • never automating systems Pattern policy forbids.

    • Cross-team work: make changes across service boundaries with the teams that own sensors, task management, approvals and the Pi web app, including in a second language (Python or Ruby on Rails) when the work requires it.

    • Requirements: translate business requirements (SOPs, approval rules, escalation paths) into clear designs, tickets and runnable systems.

    • Production support: support production, including on-call triage when an automation misbehaves.

    • Leadership (Senior): mentor other engineers, review their designs and code, and help set engineering standards for AI work on the team.

Senior Software Engineer (5-7 years)

• 3+ years of production experience in your primary backend language.

• Designs new components end to end, such as the case-tracking service, and takes ambiguous problems to a clear design.

• Owns a production service, including being on call.

• Has shipped and measured an LLM-powered feature they owned.

• Mentors engineers and reviews designs and code.

• Bachelor's degree in Computer Science, Information Technology, a related field, or equivalent professional experience.

• 5-7 years of professional software engineering experience, mostly on backend systems.

• Primary skill set (one of the following):

  • TypeScript and Python, with production experience in TypeScript (3+ years) and solid working experience in Python; or

  • TypeScript and Ruby on Rails, with production experience in either (3+ years) and solid working experience in the other.

  You should be comfortable reading and making changes in the language you use less.

  • Experience designing APIs and webhooks, including authentication and signature verification.

  • Has shipped at least one LLM-powered feature in production: tool or function calling, structured outputs, and a way to measure whether it works.

  • Comfort with AWS (ECS or containers, S3, IAM at minimum) and with Docker.

  • Production ownership,monitoring, alerting, incident response and root-cause analysis.

  • Excellent engineering practice: version control, code review, automated testing, and small, reviewable changes.

  • Strong written and spoken communication, including turning ambiguous business processes into clear technical designs.

• Experience with PostgreSQL in production: data modelling, transactions, indexing, and concurrency control (row locking, optimistic versioning, unique constraints).

• Experience with background jobs and queues (for example BullMQ, Sidekiq, Celery, SQS or Kafka), including retries, idempotency and dead-letter handling.

• Experience building agents or agent platforms: multi-step workflows, human approval steps, memory or state across runs.

• Experience building or consuming MCP servers, or similar tool-integration layers.

• Experience with LLM evaluation and observability tools such as Opik, Langfuse, LangSmith, Braintrust or Promptfoo.

• Familiarity with agent frameworks and SDKs (for example the Vercel AI SDK, OpenAI Agents SDK or LangGraph), and with LLM gateways or routers.

• Experience with durable workflow or event-driven patterns: transactional outbox, leases, Temporal, Step Functions or event sourcing.

• Experience with OpenTelemetry tracing.

• eCommerce or marketplace domain experience (Amazon, Walmart, Shopify or similar).

• Experience working productively with AI coding agents (for example Claude Code) on a large existing codebase.

• Days 1-30:

  • Learn the platform end to end: follow one automation from sensor detection to approved, completed work.

  • Ship small fixes in the agents platform and a tool server.

  • Be able to explain how a repeat signal is prevented from creating duplicate work.

  • Days 31-60:

    • Own a component, such as part of the case-tracking service, an approval integration or a set of MCP tools.

    • Add an evaluation suite that catches a real regression.

    • Close one reliability gap, for example concurrent events or stuck runs.

    • Days 61-90:

      • Lead the delivery of a new automation use case across sensors, agent, tools and approvals (Software Engineer II: own a major part of it).

      • Ship it with the evaluations, monitoring, shadow-mode run and rollback plan that make it safe to leave running.