Model Operations Lead, AI Gateway

Available Locations: Austin, TX | New York City, NY | San Francisco, CA

About AI Gateway

AI inference is becoming core infrastructure. Every serious application will need access to many models across many providers, with reliability, observability, security, cost control, and routing built in from the start. AI Gateway is Cloudflare's bet that this layer belongs at the network edge: close to users, close to compute, and simple enough that a developer can adopt it with one API change.

Developers using AI Gateway need answers to two questions. Is the model I'm calling working the way you say it does? And which model should I be using at all? You'll lead the models team that answers both.

You'll start with model operations. That's the work behind every model in the catalog: provider integrations, capability testing, pricing and metering, capacity, security reviews, docs, health, and deprecations. Sometimes a launch needs to go live within minutes of a lab's announcement. Sometimes the right call is to hold it. Cloudflare is an infrastructure company at heart, so reliability and accuracy come first; speed comes from preparation and automation.

Then you'll bring model intelligence online. That means turning what we learn from Gateway traffic into decisions for customers: capability maps, per-task rankings, auto-routing profiles that pick the right model for each request, evals and regression detection, and market intelligence developers can trust.

This is a people-management role reporting to the AI Gateway Engineering Manager. You'll hire, develop, and lead the team, and you'll stay hands-on: writing code, building agent-driven workflows, and getting into the details of a failed launch or a suspicious eval SLO. Expect to manage people, projects, model stock, and operations all at once.

About the Role

• Build and lead the models team: Hire, coach, and grow the people who run the catalog today and build its intelligence over time. Give people real ownership, give direct feedback, and grow leads who can run their own areas.
• Run model launches end-to-end: Scope integrations, coordinate testing and evals, confirm pricing, metering, capacity, and security, and get docs and announcements ready. Prepare day-zero launches in advance so they can ship in minutes.
• Keep the catalog true: Make sure what we publish (capabilities, limits, versions, prices) matches what developers actually get. Catch silent regressions, such as broken tool calling, truncated streams, quantization changes, or usage and pricing mismatches, before customers do.
• Build model intelligence: As it comes online, own the evals, capability maps, and per-task rankings behind our auto-routing profiles and model recommendations. Make sure they measure what they claim to, and connect regressions to routing decisions so customers who chose "auto" aren't the ones who discover a problem.
• Treat the catalog and intelligence as products: Have a point of view on which models belong in the catalog, which routing profiles to ship, and what market intelligence to publish. Base it on usage, quality, reliability, and cost data. Turn what you learn into roadmap input.
• Run a lightweight operating cadence: Keep one source of truth for what's in flight, who owns it, what's blocked, and what ships next. Keep Engineering, Product, Partnerships, and other stakeholders aligned and informed without status meetings.
• Automate relentlessly: Turn repeatable steps into scripts, tools, and agent workflows: endpoint validation, eval runs, catalog and pricing reconciliation, regression detection, and provider follow-ups. Manual effort per launch should