Senior AI Engineer

Zalos is building the AI action layer that sits on top of SAP, Salesforce, ServiceNow, and every other enterprise system of record your customers are trapped inside. The problem is enormous and deeply unsexy: Fortune 500 companies run on software so painful that workers toggle between applications 1,200 times a day, losing 4 hours a week just navigating screens. These systems are so welded into a company it can cost $700m to replace a core ERP (see here ).

So we're not replacing them. We're the layer across them. Zalos is the intelligent interface layer that lets enterprise teams describe what they want in plain language: "onboard this vendor, collect the docs, route approvals, set payment terms." It executes correctly across SAP, Salesforce, and ServiceNow with a full audit trail. No spelunking through 20 tabs. No tribal knowledge required.

We have paying enterprise customers, strong seed funding, and a small team that moves fast. This is a rare chance to join at the ground floor and define how AI gets work done inside the world's largest organizations.

As Zalos's Senior AI Engineer, your job is to build the AI that makes this real. You'll be designing the agent architecture that takes messy human intent and turns it into correct, auditable action against enterprise systems of record. This is the hard, important part of the product. It's entirely yours to own.

What we're building

Enterprise software was built to store data and enforce process. It was never built to be used by humans who have better things to do. The result: highly trained people wasting hours every day fighting screens instead of doing the actual work.

Zalos attacks this from three angles:

• Intent-to-action: A user describes an outcome. Zalos figures out which systems to touch, in which order, with which data, and executes it with the right approvals and a clean audit trail.
• Cross-system workflows: Event-driven chains that span SAP, Salesforce, and ServiceNow simultaneously. "If invoice posted AND variance >3%, draft an explanation and route for approval." Actions no single vendor would ever build.
• Computer-use for the long tail: Not every enterprise workflow has a reliable API. We use computer-use agents to automate the 30 to 40 percent of processes that live in screens, VDI sessions, and legacy thick clients.

Our goal is to become the trusted control plane for enterprise work: the place teams go to understand their systems, execute across them, and ship new workflows without waiting on IT.

What you'll work on

• Design and own the core agent architecture: how we take a natural language intent, decompose it into steps, call the right APIs and UI agents, handle failures gracefully, and close the loop with the user
• Build reliable, production-grade LLM pipelines covering context management, tool use, structured outputs, and multi-step reasoning across systems with real enterprise data
• Create evaluation frameworks that let us ship with confidence: evals, replay testing, regression detection, and output quality scoring against enterprise workflows
• Build computer-use agents that can navigate legacy enterprise UIs including SAP GUI, ServiceNow, and thick clients, reliably enough to trust in production
• Work directly with the founders and customers to understand where AI fails in real enterprise workflows and translate that into better systems
• Set the technical standards for how we build AI at Zalos. You'll be defining the patterns the rest of the team builds on

Must haves

We're looking for a formidable engineer. Someone who stops at nothing to get something working. Enterprise AI is unglamorous: the systems are ugly, the APIs are partial, the data is messy, and "it worked in the demo" is not good enough. You need to actually want to solve this.

• You've shipped AI-powered features that real users depend on in production. You know what breaks and how to prevent it
• Deep hands-on experience with LLM APIs, agentic patterns, and tool/function calling. Not just wrappers around OpenAI, but genuine systems-level thinking about how agents fail
• You think about reliability, latency, and cost as first-class engineering concerns, not afterthoughts
• You build evals before you ship, not after something breaks in front of a customer
• Strong Python; comfortable owning the full loop from prompt design to deployed API
• High ownership: you take something from an ambiguous problem statement to a clean, working solution with minimal hand-holding
• You're pragmatic. Startups are always on fire somewhere. You know when to let something burn and when to drop everything

Nice to haves

(Apply even if none of these apply. They're genuinely optional.)

• Fine-tuning or RLHF on domain-specific tasks
• Prior founding engineer or early-stage startup experience

The team you'll work with

You'll work directly with the founders from day one. No engineering manager in the way, no sprint planning theatre.

Why Zalos

• The market is real and enormous. The system integration market alone is $380B. Every enterprise on earth has this problem and nobody has solved it with AI yet.
• You'll work on genuinely hard AI problems. Reliable agents over messy enterprise systems, computer-use for legacy UIs, cross-system orchestration with audit trails. This is frontier applied AI, not wrapper engineering.
• Paying customers from day one. We're not pre-product. You'll see your work matter in real enterprise workflows within weeks of joining.
• Small team, high trust. You'll have a direct line to founders and customers. Your opinions shape the product.

About the interview

We move fast and respect your time. No leetcode. No whiteboard puzzles.

• 15-min intro call with a founder. We'll walk you through what we're building and what we need. You'll tell us what you've shipped.
• Take-home technical task. No trick questions. We want to see how you think, not how you memorize.
• 1-hour technical deep-dive with our CTO. Walk through your design, discuss tradeoffs, get into the weeds on production reliability.
• 3-day paid work trial. Build something real with us. This is the most predictive step we have.
• Offer. If we're aligned, you'll have an offer within a week of the trial.