Principal AI Software Engineer

PermitFlow; New York (NY); On-site; USD 180,000 - 260,000; Full time

Benefits offered by this job

Equity packages

100% Paid health, dental & vision

Company issued laptop

Home office & equipment stipend

Lunch & Dinner via UberEats

Commuter benefits

Team building events

Unlimited PTO

Job summary

PermitFlow is hiring a Staff/Principal Engineer to design and build the core AI Assistant, infrastructure, and workflow agents for processing permits and forms.

You will own the agent architecture, tool design, memory and retrieval systems, and evaluation harnesses, with 10+ years of software engineering and a history shipping AI products.

Qualifications

• 10+ years of software engineering experience.
• Experience building AI/LLM-powered products.
• Deep knowledge of agentic stack: tool calling, context, memory, retrieval.
• Ability to ship production-grade services and eval harnesses.
• Strong writing and communication.

Responsibilities

• Own the architecture of the AI assistant and infrastructure.
• Define the tools, permissions, and human-in-the-loop checks.
• Build the context layer, memory, and session models for long-running tasks.
• Stand up evaluation harnesses and the iteration loop for quality.

Skills

Software engineering

LLM-powered products

Agentic infrastructure

Tool design

Context engineering

Memory

Retrieval

Orchestration

Eval harnesses

Production-grade systems

Job description

We’re hiring a Staff/Principal Engineer to build PermitFlow’s core product AI Assistant, infrastructure, and workflow agents. Submitting a permit means comprehending construction documents, extracting project details, confirming every requirement has been met and often filling in pdf forms. Errors compound across steps and a failed submission adds real cost for our customers as projects slow. This role reports directly to our VP of Engineering. We’re hiring at Staff or Principal, with the level based on past experience and interview performance. Own the agent architecture. Design the reasoning loop, planning and decomposition strategy, state and session model, and recovery behavior for long-running tasks. Build the context layer. Retrieval, memory, and context-window management over project documents, jurisdictional requirements, prior submissions, and customer-specific conventions. Design the action surface. Define the tools the assistant can call, the permissions model around them, and the human-in-the-loop checkpoints for actions that are expensive or irreversible. Make evaluation the engine of progress. Stand up offline and online eval harnesses that measure task completion and failure modes. Build the iteration loop that lets the whole team improve agent quality on evidence. Set the bar for the org. Establish the abstractions, SDKs, and review standards other engineers build agentic features on. Stay close to the customer. Sit with permit coordinators and builders, watch the assistant fail in real conditions, and bring that back into the system design.

Benefits

• Equity packages
• 100% Paid health, dental & vision coverage
• Company issued laptop
• Home office & equipment stipend
• Lunch & Dinner provides via UberEats w/ a fully stocked kitchen
• Commuter benefits
• Team building events
• Unlimited PTO

This is a role that will require UX and technical acumen to deliver high quality agents in a genuinely complex domain, while also up-leveling our core agentic infrastructure and data flywheel. The ideal candidate will mix an exceptional history of product delivery with true experience building AI and Agentic infrastructure from the ground up. You ship quickly, instrument everything, hold a high quality bar, and stay close to the users of what you build. You are a genuinely strong software engineer, not an AI specialist who needs someone else to make it production-ready. You’ve owned services in production and you write code that other people are glad to inherit. You’ve built eval harnesses that changed what a team worked on, and you can articulate what makes an eval trustworthy. You’ve shipped an assistant or agentic product at meaningful scale. Deep, current knowledge of the agentic stack (tool calling and tool design, context engineering, memory, retrieval, orchestration patterns, etc). You operate with earned authority. You influence through demonstrated judgment, working code, and clear writing rather than title or mandate. 10+ years of software engineering experience with recent time spent building LLM-powered products.