An FAQ on careers in RL environments and post-training
Patronus AI teamAs of November 202612 min read
Patronus AI publishes this FAQ and hires for some of the roles it describes. Every claim links its source, read on Nov 19, 2026.
Work on RL environments and post-training spans research scientists and engineers who design training experiments and rewards, environment and infrastructure engineers who build sandboxes and graders, human data teams that run data projects, and domain experts who write tasks and rubrics. Frontier labs, environment startups, human data companies and research labs all hire for it. Postings read on Nov 19, 2026 listed salaries from about $144,000 to $850,000 depending on company and level.
Snapshot as of November 2026. Last updated
What jobs are there in RL environments and post-training?
Titles differ from company to company. Research scientists and research engineers design training experiments and the reward signals behind them. Environment and infrastructure engineers build the sandboxes, graders and tools those experiments run on. Human data engineers and program managers run projects that gather tasks and feedback from people. Domain experts, usually on contract, write the tasks, reference answers and rubrics. Jean-Stanislas Denain and Chris Barber wrote Epoch AI’s FAQ on RL environments (Jan 12, 2026) from interviews with 18 people at environment startups, neolabs and frontier labs. They write that building environments “mostly involves engineering skills” while creating good tasks needs domain expertise, prompting ability and product sense.
Job postings read on Nov 19, 2026 show the same split. Anthropic is hiring a Staff Software Engineer, Environments Infrastructure for an organization that “builds and maintains the infrastructure that improves Claude’s capabilities through reinforcement learning.” OpenAI’s Agent Post-Training Research posting says the team builds “the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do.” Scale AI lists a Staff Software Engineer, RL Environments. On the xAI job board, whose postings name the employer SpaceXSI, a Human Data Engineer sits “at the intersection of engineering and data operations.”
SemiAnalysis (Jan 6, 2026) describes how engineers and experts divide the work. “While software engineers construct the environment itself, often the workflows are created, described, and graded by domain specific contractors involved in the environment creation process.” For the terms used on this page, see the guide to RL environments and the FAQ on RL environments.
What does a research engineer on an RL or post-training team do?
A research engineer takes an idea for a new model skill through to a measured result. Anthropic’s Research Engineer, Code RL (Reinforcement Learning) posting (first published Jun 11, 2026) says the role will “design RL environments and coding tasks, build the reward signals and verifiers that capture what "good code" means, run training experiments on frontier models”, then work out why a model does or does not improve on a class of software engineering work. The same posting lists coding agents, code execution sandboxes, eval harnesses and verifiers as useful background.
OpenAI’s Agent Post-Training Research posting (Jun 25, 2026) calls the role “intentionally broad” and lists work that includes improving “the post-training stack, including RL, data pipelines, graders, reward signals, evals, diagnostics, and model-behavior analysis.” An xAI board posting for a Model Engineer, Post-Training & RL asks the hire to “Define what makes RL data and rewards good, including grader design and preventing reward hacking.” Reward hacking is when a model raises its score without doing the task, and Epoch AI’s interviewees named robustness against it as the top quality criterion. One environment founder told Epoch AI, “It takes many many iterations to check against reward hacking.”
Production post-training roles sit closer to the final training runs. Anthropic’s Research Engineer, Production Model Post-Training posting says the team trains base models “through the complete post-training stack to deliver the production Claude models that users interact with”, using techniques such as Constitutional AI and RLHF. It also says the role may require responding to incidents on short notice, including on weekends.
Research labs that build environments describe similar work. Patronus AI’s Member of Technical Staff - Research posting, closed as of Nov 19, 2026, listed reward design, tool simulations, agent behavior analysis and scalable oversight, and asked hires to “Train models and experiment with post-training algorithms, including GRPO and SFT.”
What does an RL environment engineer do day to day?
Mechanize, a startup that builds software engineering tasks for AI labs, describes the job in detail on its careers site. A task, it says, is “the equivalent of a take-home assessment for a coding agent”, with a prompt, room for the model to build a solution, and a grader or rubric that scores the result. One engineer owns each task from the idea to automatic grading to quality assurance, and a typical task takes about a week from ideation to submission. Mechanize writes that quality assurance takes the most time, because the grader has to become robust, fair and deterministic. Engineers there spend most of their time directing coding agents and checking their output rather than writing code by hand.
Scale AI’s Staff Software Engineer, RL Environments posting combines platform work with building environments. The platform side includes “sandboxed execution, environment packaging and versioning, rollout orchestration, trajectory capture, verifier frameworks”, and the environment side includes task suites aimed at specific capability gaps and graders “that hold up under adversarial optimization.”
Anthropic’s Environments Infrastructure posting describes an infrastructure engineer who embeds with research teams on rotation, owns the layers beneath every environment, including the agent runtime, and keeps production RL runs healthy and easy to triage. It asks candidates to “Anticipate silent failure modes and prevent them structurally through type safety, well-designed invariants, targeted testing, and refactors that reduce the room for correctness issues.”
What skills do you need to work on RL environments?
The skills depend on whether you build environments or design tasks. Epoch AI’s interviewees said task design needs domain expertise, prompting ability and product sense. One founder told Epoch AI, “Domain knowledge and expert level prompting is more important than ML skills for creating tasks.” Another of its interviewees said, “The people who are probably best at this are the people who create benchmarks that actually get used.”
Environment and infrastructure roles ask for systems engineering. Scale AI’s posting asks for 8 or more years of software engineering, strong Python, and deep experience with containers and sandboxed execution such as Docker, VMs, gVisor or Firecracker, Kubernetes or equivalent, plus hands-on work with agent loops, tool calling, MCP or eval harnesses. Among preferred skills it lists “firsthand experience with reward hacking and how to defend against it.” Anthropic’s Environments Infrastructure posting asks for deep Python expertise including static typing and async patterns, experience with stateful distributed systems, and “A habit of verification.”
Mechanize also says intuition for what models can and cannot do matters a lot and can be learned on the job, and that “reading comprehension and theory of mind are surprisingly important skills for this work”, because a good task designer predicts how a model will read a prompt and where it will take shortcuts.
Which companies hire for RL environment and post-training roles?
Frontier labs hire for this work in house. Epoch AI points to job postings at xAI and Anthropic for environment work and quotes one founder who saw “substantially more in-housing” as labs build their own data teams. On Nov 19, 2026, Anthropic’s job board listed roles such as Code RL, Cybersecurity RL, Chip Design RL and Environments Infrastructure, OpenAI’s listed Agent Post-Training and Human Data roles, and the xAI board listed post-training and RL model engineers alongside human data roles.
Environment startups also hire for this work. SemiAnalysis (Jan 6, 2026) wrote that “More than 35 companies have popped up whose goal is to do exactly this across a variety of domains”, and that almost all were seed stage with fewer than 20 employees. Mechanize describes itself as a team of around 35 people working mostly in person in San Francisco. Prime Intellect’s careers page listed 38 open roles on Nov 19, 2026, among them Research Engineer / Scientist - Reinforcement Learning and Research Engineer - RL Infrastructure.
Epoch AI names Mercor, Surge, Handshake and Turing as human data providers that now also sell RL environments. In September 2025 TechCrunch reported that Surge had spun up an internal organization to build RL environments and that Scale AI was building them too. SemiAnalysis reports that OpenAI is building an in-house human data team and that xAI has posted for AI tutors since it launched.
Research labs that build environments and publish their work hire too. Patronus AI, a behavioral research lab for AI, is one, and its researchers and engineers come from Meta Superintelligence Labs, Amazon AGI, Apple and Google (Join). As of Nov 19, 2026 its Greenhouse job board lists contract roles for domain experts in finance, healthcare, law, sales, mechanical engineering and interactive coding, plus roles such as Head of Finance and Technical Recruiter. Its open roles are on /join.
How much do RL environment and post-training jobs pay?
Many postings list a pay range. These are the ranges as each posting stated them on Nov 19, 2026, and they change often. Anthropic lists an annual salary of $500,000 to $850,000 for its Research Engineer, Code RL (Reinforcement Learning) role and $405,000 to $605,000 for its Staff Software Engineer, Environments Infrastructure role. OpenAI lists $380K to $500K for Agent Post-Training Research and $295K to $385K, one of several ranges it lists, for Software Engineer, Research - Human Data.
Scale AI lists a base salary of $252,000 to $315,000 in San Francisco, New York and Seattle for its Staff Software Engineer, RL Environments. The xAI board lists $160,000 to $250,000 for a Model Engineer, Post-Training & RL and $144,000 to $220,000 for a Human Data Engineer, with equity named as part of the package in both.
Startups publish figures too. Mechanize’s Software Engineer page lists $300K for a junior software engineer, $350K for a software engineer and $400K for a senior software engineer, each with equity and a bonus, and $150 an hour for an intern. In September 2025 TechCrunch reported that Mechanize was offering software engineers $500,000 salaries to build RL environments.
Contract work for domain experts is paid by the hour. Our FAQ on AI training work for experts lists the hourly rates postings state.
Do you need a PhD to work in RL or post-training?
Many postings do not require one. Anthropic’s Code RL and Environments Infrastructure postings list a minimum education of a “Bachelor’s degree or an equivalent combination of education, training, and/or experience.” The xAI board’s Human Data Engineer posting asks for a bachelor’s degree in engineering, computer science or a STEM field, or 4 or more years of experience instead. Its Model Engineer, Post-Training & RL posting lists production-quality code in Python, PyTorch, JAX, Rust or C++ as the basic qualification and says prior post-training work is a plus “but relevant experience is not required.”
Research scientist roles ask for more. Patronus AI’s closed Member of Technical Staff - Research posting asked for “An MS or PhD in Computer Science, Machine Learning, Statistics, Mathematics, or a related quantitative field” and experience running independent research. Mechanize hires junior software engineers and interns for task building.
For task design, Epoch AI writes that having a lot of experience interacting with frontier models is key. One neolab researcher said, “You don’t necessarily need to be an AI researcher, but perhaps a very heavy Claude Code user, a prompt whisperer like Riley Goodside, can be better at figuring out what the frontier is than an AI researcher.”
What do domain experts do on RL environment teams?
Experts supply the knowledge a task needs to be realistic and gradable. SemiAnalysis writes that finance professionals help define financial tasks, doctors medical ones and lawyers legal ones, and that “Contractors design tasks, write expected solutions, specify reward signals, and in some cases, grade model solutions.” It adds that such contracts typically last at least a quarter and can be part time or full time.
Postings describe the same tasks. An xAI board posting for an AI Tutor, Medicine - Temporary asks medical experts to create and solve medical problems in their area, “writing detailed reference answers and critically evaluating model responses.” Scale AI’s Subject Matter Expert posting focuses on audits and asks experts to “Define and refine quality benchmarks, scoring rubrics, and evaluation criteria based on project needs.” Patronus AI’s finance expert posting asks experts to build tasks from a scenario, including the working files, the request, their own answer and how it should be graded, and to review tasks other experts build.
Our FAQ on AI training work for experts covers fields, posted pay, contracts and how to spot fake postings.
What is a human data team at an AI lab?
A human data team runs the projects in which people write, rate and correct the examples a model learns from. OpenAI’s Human Data posting says its engineering team “creates the systems that enable scalable, high-quality human feedback”, and that engineers build systems for feedback collection, data labeling and evaluation pipelines. OpenAI’s job board on Nov 19, 2026 also listed a Data Program Manager, Human Data and a Strategic Sourcing Lead, Professional Services – Human Data.
The xAI board’s Human Data Engineer posting describes partnering with model teams to design data projects and building “tools and systems to measure data effectiveness (data yield, eval lift, usage)”. It asks for 2 or more years building production tools or data systems, such as annotation, data quality, evaluation or human-in-the-loop tooling.
Labs also buy this work from outside firms. SemiAnalysis reports that firms such as Mercor, Handshake, Surge and Aboda.ai hire experts across professional domains for the labs, and that OpenAI runs a private internal platform for contractors named Feather. Comparisons of these vendors are in our Mercor alternatives and Scale AI alternatives roundups.
About this FAQ
Patronus AI publishes this FAQ. It builds RL environments and hires domain experts, so it appears as one example among the employers described. The format follows Epoch AI’s FAQ on RL environments, which this page cites.
Job postings change often. Every posting cited here was read on Nov 19, 2026, and the Sources list gives each one with its date. To report an error, write to us through the contact page.
Sources
- 1 Jean-Stanislas Denain and Chris Barber, “An FAQ on reinforcement learning environments”, Epoch AI Gradient Updates, Jan 12, 2026
- 2 AJ and Dylan Patel, “RL Environments and RL for Science: Data Foundries and Multi-Agent Architectures”, SemiAnalysis, Jan 6, 2026 (free portion read)
- 3 Maxwell Zeff, “Silicon Valley bets big on ‘environments’ to train AI agents”, TechCrunch, Sep 16, 2025 (updated Sep 21, 2025)
- 4 Anthropic, job board, read Nov 19, 2026
- 5 OpenAI, job board, read Nov 19, 2026
- 6 xAI job board (postings name the employer SpaceXSI), read Nov 19, 2026
- 7 Anthropic, “Research Engineer, Code RL (Reinforcement Learning)”, job posting, first published Jun 11, 2026, read Nov 19, 2026
- 8 Anthropic, “Staff Software Engineer, Environments Infrastructure”, job posting, first published Jul 24, 2026, read Nov 19, 2026
- 9 Anthropic, “Research Engineer, Production Model Post-Training”, job posting (Zürich), first published Feb 12, 2026, read Nov 19, 2026
- 10 OpenAI, “Agent Post-Training Research”, job posting, published Jun 25, 2026, read Nov 19, 2026
- 11 OpenAI, “Software Engineer, Research - Human Data”, job posting, published Jan 23, 2026, read Nov 19, 2026
- 12 SpaceXSI (xAI job board), “Model Engineer, Post-Training & RL”, job posting, first published Apr 28, 2026, read Nov 19, 2026
- 13 SpaceXSI (xAI job board), “Human Data Engineer”, job posting, first published Aug 5, 2026, read Nov 19, 2026
- 14 SpaceXSI (xAI job board), “AI Tutor, Medicine - Temporary”, job posting, read Nov 19, 2026
- 15 Scale AI, “Staff Software Engineer, RL Environments”, job posting, first published Sep 2, 2026, read Nov 19, 2026
- 16 Scale AI, “Subject Matter Expert”, job posting, first published Jun 9, 2026, read Nov 19, 2026
- 17 Mechanize, “What working at Mechanize is like”, read Nov 19, 2026
- 18 Mechanize, “Software Engineer” application page, read Nov 19, 2026
- 19 Prime Intellect, Careers, read Nov 19, 2026
- 20 Patronus AI, Join, read Nov 19, 2026
- 21 Patronus AI open roles, Greenhouse job board API, read Nov 19, 2026
- 22 Patronus AI, “Member of Technical Staff - Research”, job posting, closed as of Nov 19, 2026
- 23 Patronus AI, “Expert: Finance (Contract, Remote)”, job posting, read Nov 19, 2026
Patronus AI · November 2026 snapshot · More resources