AI Agent Engineer — open roles, skills, and who's hiring
There are 371 open agent engineer roles across 47 companies tracked by the Agentic Ready Jobs Index, as of 12 June 2026. That makes agent engineering the largest category in the index — 43% of all 866 agentic roles tracked — and 205 of those 371 postings carry "forward deployed" in the title. Only 32 of the 371 (9%) are remote, which is low for software engineering and tells you something about how hands-on this work still is.
What an agent engineer is
An agent engineer builds software systems where a language model decides what to do next: which tool to call, what to ask the user, when to stop. The job sits one layer above model work and one layer below product work. You are not training models, and you are not only shipping features — you are designing the loop in which a model plans, acts, observes the result, and acts again, and making that loop reliable enough to put in front of customers.
In practice the work splits into a few recurring problems. Orchestration: sequencing model calls, tool calls, and human checkpoints, which is why titles like "Staff Software Engineer, Agent Orchestration" and "Software Engineer, Agent Harness" appear in the current postings. Tooling: giving the agent safe, well-described interfaces to real systems. Evaluation: building the test harnesses that tell you whether a change made the agent better or just different. And failure handling: agents fail in open-ended ways, so a large share of the job is bounding what a wrong action can cost.
The title is young and unsettled. The same work is posted as "Software Engineer, Agents", "Agent Engineer", "Frontier Agents Engineer", and — in over half the category — "Forward Deployed Engineer", where the agent-building happens inside a customer's environment. If you are searching for this job, search the work, not the title.
Skills and tools that appear in real postings
Language-wise this is ordinary production engineering: the postings are software engineering roles first, with Python and TypeScript the default assumption at most of the companies listed below.
How to break in
The most common path in the current data is sideways from production software engineering: 241 of the 371 postings (65%) are mid-level, so companies are hiring engineers who can learn the agent layer rather than demanding years of agent experience nobody has. A public project where an agent does real multi-step work — with an eval suite attached — covers what most interviews probe.
Adjacent roles worth comparing: AI engineer if you want to sit closer to the model, agent ops engineer if you prefer the infrastructure under the agents, forward deployed engineer if you want the customer-facing variant, and AI governance lead if the permission-boundary part of the job interests you more than the building part. The full dataset is on the Agentic AI Jobs Index and the tracker.
Skills appearing in real postings
Hiring for this role right now
- Salesforce 137 roles San Francisco Careers ↗
- OpenAI 78 roles San Francisco Careers ↗
- Databricks 70 roles San Francisco Careers ↗
- Citi 65 roles New York Careers ↗
- LangChain 61 roles San Francisco Careers ↗
Live from the Agentic AI Jobs Index, updated 16 June 2026.
Salary
None of the 371 agent engineering postings in the 12 June 2026 snapshot disclosed a salary range, and levels.fyi does not yet track "agent engineer" as a distinct title. The closest public benchmark is the ML/AI software engineer category on levels.fyi, with a median total compensation of $243,000 — self-reported and skewed toward large US tech companies, so treat it as a ceiling-ish reference, not a market rate.
Sources: levels.fyi — ML/AI software engineer