Software Engineer, Agentic Tooling, Tesla AI

Tesla AI's Agentic Tooling team builds the harness and skill ecosystem for the AI org powering FSD and Optimus. Our agents already do real engineering work day-to-day for engineers across the org. We own the customized harness end-to-end: the orchestration around model backends, the tool ecosystem agents call into, the session and memory layer, the observability stack, and the evaluation system that gates every prompt, skill, and model upgrade. You'll work across the full surface of the team, extending the harness, shipping new skills, building evaluations, owning observability, and picking up whatever is most leveraged each quarter. Models and skills evolve weekly, so a fast iteration loop and honest measurement of what you ship are non-negotiable.

Responsibilities

Harness development: Extend the agent runtime and tool ecosystem end-to-end. Add new tools, integrations, and backends without regressing existing skills, and keep the platform easy to extend as the ecosystem evolves Skill development: Ship new skills that solve real workflows for engineers across the AI org, from one-off automations to the agent capabilities that get reused across the fleet Evaluation: Build and operate the evaluation framework that catches regressions in prompts, skills, harness changes, and model upgrades before they reach users Observability and reliability: Own session-level observability across our tracing, logging, and metrics stack. Build the dashboards and alerting that tell us when an agent is broken before users do Production infrastructure: Operate our production agent services on Kubernetes, async backend services, durable storage for session memory and embeddings, and the public-facing APIs that route work to them Contribute to architectural decisions: With a focus on security, scalability, and reliability, especially around credential isolation and the seams between our harness, the model runtime, and the sandboxing layer Embed with users and ship cross-team, end-to-end: This is a collaboration-heavy role. You'll sit with engineers across the AI org to understand their workflows, identify the moments where an agent could remove toil, and drive the solution from the first conversation through prototype, rollout, and hand-off. Many of our highest-impact tools start as zero-to-one builds where you step in, learn enough of a stakeholder's domain to understand what "done" looks like, and ship the first version yourself

Requirements

Proficiency with Python; Go is a plus Strong foundation in concurrent, async, and distributed systems, orchestration, durable execution, and state across services Strong software-design fundamentals, SOLID, separation of concerns, clean abstractions and interfaces, and composition over inheritance Hands-on experience building with agents (tools, skills, harnesses, or agent-powered side projects) Open-ended problems and product instincts: You enjoy talking to engineers across teams, can convert a vague ask into a shipped solution end-to-end, and bias toward a usable v1 that people adopt over a "complete" solution that takes months to set up Adaptability and curiosity in a fast-paced space: Agentic tooling moves weekly. You stay current on new models, frameworks, and agent patterns as a matter of habit, and you can re-platform your own work when the right answer changes underneath you Familiarity with systems and infrastructure -- operating systems, storage, relational and vector databases, and container orchestration (Docker, Kubernetes), is a plus Experience across the agentic dev-tooling ecosystem is a plus, the Claude Code SDK, ACP (Agent Client Protocol), or opencode; building or operating evaluation pipelines (offline replay, ground-truth scoring, CI gating); or developer-tools work where the customer is another engineer Experience with sandboxing and application security (sandboxed code execution, container isolation, capability-based access control, credential isolation, secrets management) is a plus