Founding AI Engineer — Agent Runtime

About Pareto Agent Most AI systems generate text. We’re building one that makes decisions . Pareto Agent is a policy-driven runtime that executes high-stakes commercial workflows—where every action either protects revenue or gives it away. The future of B2B sales isn’t more reps -it is smarter systems . We’re building the execution layer that replaces manual, inconsistent decision-making with deterministic, system-driven outcomes . Inspired by the Pareto Principle (80/20), we focus on the small number of decisions that drive the majority of outcomes—and build systems that execute them with precision. Headquartered in San Francisco, the company is founded by serial entrepreneurs who have successfully scaled multiple B2B companies and has secured over $3.5M in funding from their previous investors. The Role As our Founding AI Engineer - Agent Runtime , you will own the execution system behind an autonomous AI sales agent. This is not a typical LLM application. You’ll be building a policy-driven execution system where model outputs are constrained, evaluated, and enforced by a deterministic runtime. Reporting directly to the CTO/Co-Founder, you will play a critical role in shaping both the technical architecture and product direction. The systems you build will negotiate real contracts, protect real revenue, and operate within real-world constraints. The quality of your engineering is the difference between an AI that closes deals and one that gives away margin. What You’ll Do Own the end-to-end execution system — including the agent pipeline and the event-driven runtime that governs lifecycle, state, and policy enforcement Design and evolve a multi-stage agent pipeline (intent classification, context assembly, reasoning, response generation) as a cohesive, testable system Build and maintain a robust evaluation framework — defining correctness and catching regressions before they reach customers Work within a structured rules and policy system — including constraints, escalation logic, and commercial guardrails Design systems that are safe by construction , ensuring the agent operates within pricing, legal, and approval boundaries at the architecture level Architect context assembly — determining what to include, retrieve, compress, or discard as complexity scales Build instrumentation and feedback loops so every interaction improves system performance over time Make the agent configurable and increasingly self-sufficient — start from the rules, guardrails, and communication guidelines customers define today; build the instrumentation and feedback loops that reduce how much explicit configuration is needed tomorrow Lay the foundation for a self-improving system , where outcomes drive better models, smarter context selection, and improved decision-making Engage with early customers to validate assumptions and translate real-world usage into product direction Who You Are You’ve built production LLM systems where “it usually works” wasn’t good enough You have strong judgment on when to rely on models vs. when to enforce constraints You think in types and systems — production experience in TypeScript, Go, Rust, or a comparable typed language You have experience designing evaluation frameworks and care deeply about correctness You understand tradeoffs in retrieval, context management, and multi-turn reasoning You can diagnose whether a failure is a prompt issue, system design issue, or task misfit for LLMs You’re highly self-directed and comfortable operating in ambiguous, early-stage environments You're pushing toward agent-first development — you treat coding agents as an execution layer and invest in the scaffolding, feedback loops, and environment design that let them do reliable, high-throughput work Nice to Have: Background in ML or data systems Experience in B2B SaaS, sales tech, or revenue workflows Our Culture & How We Work Customer First: We build for outcomes, not output. If it doesn't make our product a must-have for B2B e