Agent ops engineer

Agent Ops Engineer — open roles, skills, and who's hiring

There are 78 open agent ops and infrastructure roles across 20 companies tracked by the Agentic Ready Jobs Index, as of 12 June 2026. Inference is the centre of gravity: 40 of the 78 postings (51%) carry "inference" in the title, and another 20 name a platform. Only 5 of the 78 (6%) are remote — among the lowest remote shares in the index.

Open roles160
Companies51
Remote share8%

What an agent ops engineer is

An agent ops engineer runs the infrastructure that AI systems — models and the agents built on them — depend on in production. Where an agent engineer designs the loop an agent executes, the ops engineer makes sure the loop has something reliable to execute on: model serving that meets latency targets, platforms that let product teams deploy without reinventing infrastructure, routing that sends each request to the right model at an acceptable cost, and the monitoring that says what every agent did and what it spent.

The current postings split into two jobs that share a category. The larger one is inference engineering: serving models fast and cheaply, with titles like "Performance Engineer, Inference Systems", "Software Engineer, Model Routing & Inference", "Site Reliability Engineer, Inference Infrastructure", and niche variants down to "Inference – AMD GPU Enablement" and on-device transformers. This is systems performance work — kernels, GPUs, schedulers — applied to models. The smaller one is platform engineering: "Staff Software Engineer, AI Platform" and its variants, building the internal layer through which a company's teams ship AI features. The platform job looks more like conventional infrastructure engineering and is the more common entry point of the two.

"Agent ops" as a discipline — operating fleets of agents, with identity, budgets, kill switches, and audit trails per agent — is younger than either, and mostly appears inside the platform postings rather than under its own title. Expect that to change as agent fleets grow past what a team can supervise informally; the operational tooling category (Prefactor is one example of a vendor in it) exists because this gap is now visible.

Skills and tools that appear in real postings

Seniority skews experienced: 16 of 78 postings (21%) are staff-plus and 14 more are management.

How to break in

The platform side is the accessible entrance: if you have run production infrastructure — SRE, platform engineering, large-scale backend — the AI platform roles ask you to apply that practice to a new workload, and 35 of the 78 postings are mid-level. The inference-performance side has a steeper ramp (GPU programming, serving frameworks, performance work) but less competition, because systems-performance engineers are scarcer than platform engineers. Moving from within — taking ownership of your current employer's model serving or LLM gateway — remains the most common route in practice.

Adjacent roles: agent engineer sits one layer up, AI engineer builds on the platforms you would run, and AI governance lead consumes the audit trails and controls you would build. The full dataset is on the Agentic AI Jobs Index.

Skills appearing in real postings

Inference performanceGPU enablementModel routingAI platform engineeringKubernetesObservabilityReliability engineeringGPU capacity planning

Hiring for this role right now

Live from the Agentic AI Jobs Index, updated 16 June 2026.

Salary

None of the 78 postings in the 12 June 2026 snapshot disclosed a range. The nearest public benchmarks are imperfect: levels.fyi's machine learning engineer title shows a median of $270,000 (self-reported, large-company skew), and a staffing-firm analysis of MLOps ranges spans $90,000 to $257,000 — a spread wide enough to confirm only that the title covers several different jobs. Disclosed figures for inference-specific roles are rare; the dataset will publish them when postings include them.

Sources: levels.fyi — machine learning engineer · Kore1 MLOps salary guide

Common questions

Frequently asked questions

How many agent ops and LLMOps jobs are open right now?

78 open agent ops and infrastructure roles across 20 companies, as of 12 June 2026, per the Agentic Ready Jobs Index. About half (40) are inference-focused.

What is the difference between MLOps and agent ops?

MLOps operates the model lifecycle — training, deployment, serving. Agent ops extends that to running agents as operational actors: per-agent identity, permissions, spend limits, and audit trails on top of the serving layer. In the current market the second job is mostly hired under platform titles; only the first has an established name.

Do agent ops roles require GPU experience?

About half do. The 40 inference-titled postings generally expect serving and performance work close to the hardware; the 20 platform-titled postings mostly do not, and lean on conventional infrastructure skills instead.

Hiring for this role — or trying to become the organisation that can use it? See where you stand.

Take the assessment →