The next frontier of applied AI isn’t another chat window embedded in a system of record. It’s systems that can perceive, reason about, and act on the physical world. Restaurants are a $1.5T industry that runs almost entirely on tribal knowledge and guesswork, and where the data that matters has never before been captured cleanly.
At Marble, we’re building the first agentic operating system for physical industries, starting with multi-unit restaurant groups.
We've spent decades in this industry and have shipped AI/ML systems at scale for some of the largest institutions on the planet. That means we speak the customer's language and are building workflows around processes we already understand deeply. In less than four months, we signed 150+ restaurants across the US, Canada, and Europe, and have grown 25% week over week for three months and counting. We're backed by Y Combinator and a host of incredible investors, angels, and operators.
The Role
We're looking for a product-minded founding engineer who will challenge us, push the boundaries of what's possible, and shape how operators experience our product. Engineers at Marble are not ticket-takers: you'll drive the product roadmap, speak with customers, sell the product, and have a voice in strategic decisions from day one.
We have a strong preference for individuals with a founder tilt: people who have founded something, want to found something, or simply can't stop building.
What you'll work on:
Build the full lifecycle: Architect and ship customer-facing functionality from 0-to-1, from the agent that drafts a purchase order to the interface a manager approves it in.
Own the agent harness: Build the systems that let models reason over operational data, use tools, and carry work through to completion. Own context management, orchestration, persistent state, and human approval flows for agents making purchasing and inventory decisions with real dollars attached.
Solve hard data problems: Backend APIs, integrations (POS, accounting, payroll), background jobs, and pipelines that turn invoices, recipes, vendor catalogs, and stock counts into clean ground truth.
Make agents dependable: Build evaluations from real operator workflows, trace failures across model and tool interactions, and improve how agents recover or escalate when they get stuck. Use production evidence to choose models and improve task completion, latency, and cost.
Drive user experience: Work across backend services and React interfaces to deliver complete features. Bridge the gap between "working code" and a product operators want to open every morning.
Learn from production: Take customer feedback and real-world behavior and turn it into improvements in accuracy, usability, speed, and cost.
Requirements
Can be in office 5 days/week in New York City
A technical spike: Deep, hands-on ability in at least one of: LLM/agentic applications, backend and data systems, or product engineering in React. Strong fundamentals in Python or JavaScript/TypeScript, with willingness to become productive in both.
Evidence of building: You've completed something substantial enough to discuss its architecture, implementation choices, limitations, and what you'd do differently. Personal projects, internships, research implementations, and open-source contributions all count.
Production agent engineering: You’ve shipped agents that use tools and take actions in production. You’ve built and debugged the harness around them: context management, tool execution, state, failure recovery, and evaluations. You’ve worked across model families and providers, and can explain how their capabilities and failure modes shaped your architecture.
Cloud infrastructure: Experience building and operating production systems on GCP, AWS, Azure, or similar platforms, including compute, databases, messaging, storage, and access management.
Agency and ownership: You thrive when you define the solution rather than implement a spec. You'd rather argue for a better approach than quietly build the wrong one. You take initiative, follow through, ask for help when needed, and communicate progress and blockers clearly.
Product intuition: You want to understand the problem behind a request. You care about the "why" as much as the "how."
Engineering habits: Git, readable code, tests that validate changes, systematic debugging. You use AI coding tools to move faster while still understanding, reviewing, and testing what they produce.
Attention to messy details: You notice missing data, duplicate records, unexpected inputs, and confusing UX.
Nice to Haves
You don't need all of these. Candidates can be stronger in some areas and grow into others.
Agent frameworks: Experience with LangGraph, LangChain, or similar orchestration tools.
Model and tool judgment: You understand how model capabilities, tool design, and context interact. You can explain when to use an agent loop, when a deterministic workflow is sufficient, and how you evaluate a new model or framework before putting it into production.
Data modeling: SQL and/or document databases; handling incomplete or inconsistent inputs
Product engineering: React and TypeScript; thoughtful loading, error, and success states
AI quality and ops: evals, logging/tracing, latency, token cost, permissions, safe handling of model-generated actions. LangSmith is a plus
Founder tilt: You've founded something, want to, or build things obsessively on your own time. You're comfortable talking to customers, selling what you built, and owning outcomes rather than tasks.
Interest in document extraction/OCR, computer vision, forecasting, procurement, logistics, or restaurant operations
Worked at an early-stage startup
Why you should join Marble
The frontier of one of the largest industries on the planet: Restaurants are a $1.5T industry that has never had software built for how it actually runs.
Rapid growth: 150+ restaurants in under four months, 25% week over week for three months and counting. What you ship this week runs in real kitchens next week.
High agency: Engineers at Marble are product builders, not ticket-takers. You'll drive the roadmap, sit in customer calls, and have a voice in strategic decisions from day one.
Outstanding performance is rewarded early and often: We don't wait for annual cycles. Compensation, equity, and scope grow with your impact, not your tenure.
Ownership: We find the intersection between what excites you and what the business needs, and hand you ownership of it.
More About Us
A restaurant operation generates enormous amounts of information every minute - what's on the shelf, what came off the truck, what got thrown out, what ran short on the line - but none of it exists in a form any database was designed to store.
We're solving three hard problems that sit on top of each other:
Ground truth capture. A multimodal input layer - computer vision, voice AI, document and web ingestion, third-party integration - that captures what's actually happening inside the operation.
Prediction on stochastic data. Once the ground truth is captured, the question becomes what happens next. Demand moves with the weather, a game schedule, a promo two blocks over, a catering order. Forecasting here isn't a curve fit over clean history, it's continuous inference over noisy and shifting signals across dozens of locations.
Long horizon agents that act. Agents take the ground truth and the forecast and make decisions with real money attached: what to order, how much to prep, who to schedule, which invoice to dispute. They run every day, across every location, and automate tens of thousands of manual workflows.
Solve all three, and you've transformed the most essential industry on the planet from one that has run on guesswork for millennia to one that operates with the discipline of a modern fulfillment center.
Source: Y Combinator job 109611. Listed on YC when retrieved 2026-11-02.
