Founding AI Engineer — Agent Runtime

Company Description

Pareto Agent - VC-backed AI startup building a policy-driven runtime for high-stakes commercial workflows.

Job Description

Join Pareto Agent as the first engineering hire to build the execution layer for autonomous B2B sales agents. Unlike standard LLM wrappers, you will architect a deterministic, policy-driven runtime that governs agent actions in real-world negotiations. Reporting to a serial-entrepreneur CTO, you'll ensure every AI decision protects revenue and follows strict guardrails.

Location

San Francisco, USA

Why this role is remarkable:

• Founding opportunity as the first engineer, offering significant 1.5% equity and direct influence on the technical architecture of a $3.5M-funded startup.
• Work alongside serial founders Evan Liang (ex-LeanData CEO) and Maurizio Greco (20+ year CTO) who have successfully scaled multiple B2B companies.
• Tackle deep systems engineering challenges by moving beyond simple chat interfaces to build a self-improving, safe-by-construction execution system for high-stakes commerce.

What you will do:

• Design and own the multi-stage agent pipeline, including intent classification, context assembly, and reasoning loops to enable reliable autonomous decision-making.
• Build a robust event-driven runtime and evaluation framework to enforce commercial policies and catch regressions before they impact customers.
• Architect sophisticated context management systems that determine how to retrieve, compress, and utilize data as agent complexity scales across various industries.

The ideal candidate:

• Proven experience building production-grade LLM systems where reliability was mission-critical, moving beyond "it usually works" to deterministic outcomes.
• Strong systems thinking with professional proficiency in typed languages such as TypeScript, Go, or Rust for building scalable, type-safe infrastructure.
• Self-directed engineer comfortable in early-stage environments who can diagnose whether failures stem from prompts, system design, or LLM limitations.