Engineering Manager, Inference Infrastructure

About Anthropic

Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.

About the role

Every request that hits Claude — from claude.ai, the API, our cloud partners, or internal research — depends on a set of decisions made before it ever reaches a model: where each request should be served and how much capacity each model needs right now. Getting those decisions right is crucial to satisfying throughput, reliability, and latency constraints. This group builds the control plane that makes those decisions for Anthropic's inference fleet and own the inference request path.

This is a deeply technical group. The engineers here design placement and load-balancing algorithms, build quantitative models of demand, capacity, and system performance, improve latency across kernel, network, and framework boundaries, and reason carefully about how a change to the fleet ripples through everything that depends on it.

You'll lead a strong group of ML platform, infrastructure, and distributed-systems engineers working alongside the teams that build our ML internals and cloud infrastructure. You need enough systems depth to make architectural calls, hire people who go deep, and see when a proposed change will ripple across the fleet. You're accountable for the health of the whole path from request to model: its efficiency, its reliability, and how well it evolves as models, hardware, and clouds change underneath it.

Key responsibilities

- Own the technical roadmap for how the inference fleet is coordinated — where traffic goes, where capacity lives, how caches are placed, how fast the system reacts to demand, and the protocols that keep the control plane and the inference engines in sync

- Partner with the product, inference engine, performance, and capacity teams to identify throughput, latency, utilization, and cost wins, then turn those into shipped improvements with measurable results

- Build the group's habit of quantitative modeling: claim a win only when you can measure it, and know before you ship what the expected effect is

- Set technical strategy for how the control plane evolves across heterogeneous hardware, across multiple cloud providers, and across all our serving surfaces

- Run the group's operational backbone — on-call rotations, incident response, postmortem review, deploy safety — so the teams can ship aggressively without the system becoming fragile

- Create clarity at a seam: this group sits between the API surface, the inference engines, capacity planning, and the cloud deployment teams