Agent Post-training, Context Research

OpenAI · AI Frontier · San Francisco, CA · Research

Research role focused on training frontier agents, specifically improving post-training capabilities and scaling compute on context. Responsibilities include designing experiments, owning improvements to the post-training stack (RL, data pipelines, graders, reward signals, evals), building evals and environments, partnering with product teams, working on early-training and alignment interventions, and debugging model failures. The role emphasizes hands-on experience with LLMs, RL, RLHF/RLAIF, post-training, evals, and production ML systems, with a focus on product impact and model behavior.

What you'd actually do

Design and run experiments that improve scaling of compute on context.
Own end-to-end improvements to the post-training stack, including RL, data pipelines, graders, reward signals, evals, diagnostics, and model-behavior analysis.
Build evals and environments that expose the next set of model failures, then turn those failures into training data, product fixes, or new research directions.
Partner with Codex and ChatGPT product teams to understand what users need and translate product signal into model improvements.
Work on early-training and alignment interventions, including data mixtures, objectives, synthetic data, and eval loops that shape downstream agent behavior.

Skills

Required

strong technical fundamentals in machine learning, software engineering, systems, statistics, or a related field
learn quickly across the parts you have not worked in before
hands-on experience with LLMs, RL, RLHF/RLAIF, post-training, evals, graders, synthetic data, model training, coding agents, tool-using agents, or production ML systems
excited by open-ended problems where the path is unclear, the signal is noisy, and the right answer requires both research taste and engineering execution
care about product impact and model behavior, not just benchmark movement
opinions about what makes an agent useful, reliable, honest, tasteful, and easy to work with
can move from a vague behavioral problem to a concrete experiment: define the hypothesis, build the pipeline, run the model, analyze the result, and decide what to do next
comfortable working across research, product, infrastructure, data, evals, and safety boundaries
can communicate clearly with each group
like building load-bearing systems and processes when that is what the team needs
want to train and ship the models that make agents genuinely useful for developers, enterprises, researchers, and everyday users

Nice to have

multi-agent coordination
long-horizon execution
factuality
instruction following
calibrated reasoning
taste
Codex Chronicle
Codex and ChatGPT product teams
model training
product infrastructure
production agent harness
multi-agent systems
training directly against production-like environments

What the JD emphasized

frontier agents
post-training
scale compute spent on context
frontier training stack
iterative deployment
major model runs
product interface
product signal into model improvements
early-training and alignment interventions
agent behavior
production agent harness
multi-agent systems
training directly against production-like environments
debug hard failures in shipped or near-shipped models
LLMs
RL
RLHF/RLAIF
post-training
evals
graders
synthetic data
model training
coding agents
tool-using agents
production ML systems
product impact
model behavior
useful, reliable, honest, tasteful, and easy to work with
vague behavioral problem to a concrete experiment
research taste and engineering execution
move from a vague behavioral problem to a concrete experiment
build the pipeline, run the model, analyze the result, and decide what to do next
communicate clearly with each group
build load-bearing systems and processes
train and ship the models that make agents genuinely useful

Other signals

frontier agents
training the models behind our agents
build the data, environments, graders, training methods, and feedback loops
scale compute spent on context
frontier training stack
iterative deployment
major model runs
product interface
product signal into model improvements
early-training and alignment interventions
agent behavior
production agent harness
multi-agent systems
training directly against production-like environments
debug hard failures in shipped or near-shipped models
LLMs
RL
RLHF/RLAIF
post-training
evals
graders
synthetic data
model training
coding agents
tool-using agents
production ML systems
product impact
model behavior
useful, reliable, honest, tasteful, and easy to work with
vague behavioral problem to a concrete experiment
research taste and engineering execution
move from a vague behavioral problem to a concrete experiment
build the pipeline, run the model, analyze the result, and decide what to do next
communicate clearly with each group
build load-bearing systems and processes
train and ship the models that make agents genuinely useful

Read full job description

About the Team

The Agent Post-Training team creates the frontier agents OpenAI ships to the world. We are training the models behind our agents in Codex, ChatGPT, the API, and other frontier products: persistent, proactive intelligence that can operate computers, collaborate with people and other agents, and expand what people and organizations can imagine, attempt, and achieve.

We define what the next generation of agents should be able to do, build the training signal that teaches those abilities, and run the experiments that make them real. Our work spans coding, tool use, computer use, multi-agent coordination, long-horizon execution, factuality, instruction following, calibrated reasoning, and taste.

Our team is where new model capabilities get made. We build the data, environments, graders, training methods, and feedback loops that shape what OpenAI's next agents can do, then carry those capabilities through major training runs and into the products people use.

About the Role

We believe that the final enabler for AGI is spending compute on context. As a Context Researcher on Agent Post-Training, you will scale compute spent on context. You will get to work in our frontier training stack on enabling the next paradigm of model training with a clear product interface for iterative deployment (Codex Chronicle). You will work with researchers, engineers, product teams, infrastructure teams, and safety/alignment partners to decide what should go into major model runs, measure whether it worked, and ship improvements into products used by real people. This is a high-agency role for people who want their work to land directly in frontier models.

In this role, you might

Design and run experiments that improve scaling of compute on context.
Own end-to-end improvements to the post-training stack, including RL, data pipelines, graders, reward signals, evals, diagnostics, and model-behavior analysis.
Build evals and environments that expose the next set of model failures, then turn those failures into training data, product fixes, or new research directions.
Partner with Codex and ChatGPT product teams to understand what users need and translate product signal into model improvements.
Work on early-training and alignment interventions, including data mixtures, objectives, synthetic data, and eval loops that shape downstream agent behavior.
Help decide which integrations, capabilities, and fixes are ready for inclusion in major model runs.
Improve the machinery for large-scale training and launch: experiment velocity, reliability, observability, reproducibility, cost, latency, and production readiness.
Take on cross-functional projects that touch model training, product infrastructure, and the production agent harness, such as multi-agent systems or training directly against production-like environments.
Debug hard failures in shipped or near-shipped models and turn messy qualitative behavior into concrete hypotheses, experiments, and fixes.

You might thrive in this role if you

Have strong technical fundamentals in machine learning, software engineering, systems, statistics, or a related field, and can learn quickly across the parts you have not worked in before.
Have hands-on experience with LLMs, RL, RLHF/RLAIF, post-training, evals, graders, synthetic data, model training, coding agents, tool-using agents, or production ML systems.
Are excited by open-ended problems where the path is unclear, the signal is noisy, and the right answer requires both research taste and engineering execution.
Care about product impact and model behavior, not just benchmark movement. You have opinions about what makes an agent useful, reliable, honest, tasteful, and easy to work with.
Can move from a vague behavioral problem to a concrete experiment: define the hypothesis, build the pipeline, run the model, analyze the result, and decide what to do next.
Are comfortable working across research, product, infrastructure, data, evals, and safety boundaries, and can communicate clearly with each group.
Like building load-bearing systems and processes when that is what the team needs, even if the work is not glamorous.
Want to train and ship the models that make agents genuinely useful for developers, enterprises, researchers, and everyday users.

About OpenAI

OpenAI is an AI research and deployment company dedicated to ensuring that general-purpose artificial intelligence benefits all of humanity. We push the boundaries of the capabilities of AI systems and seek to safely deploy them to the world through our products. AI is an extremely powerful tool that must be created with safety and human needs at its core, and to achieve our mission, we must encompass and value the many different perspectives, voices, and experiences that form the full spectrum of humanity.

We are an equal opportunity employer, and we do not discriminate on the basis of race, religion, color, national origin, sex, sexual orientation, age, veteran status, disability, genetic information, or other applicable legally protected characteristic.

For additional information, please see OpenAI’s Affirmative Action and Equal Employment Opportunity Policy Statement.

Background checks for applicants will be administered in accordance with applicable law, and qualified applicants with arrest or conviction records will be considered for employment consistent with those laws, including the San Francisco Fair Chance Ordinance, the Los Angeles County Fair Chance Ordinance for Employers, and the California Fair Chance Act, for US-based candidates. For unincorporated Los Angeles County workers: we reasonably believe that criminal history may have a direct, adverse and negative relationship with the following job duties, potentially resulting in the withdrawal of a conditional offer of employment: protect computer hardware entrusted to you from theft, loss or damage; return all computer hardware in your possession (including the data contained therein) upon termination of employment or end of assignment; and maintain the confidentiality of proprietary, confidential, and non-public information. In addition, job duties require access to secure and protected information technology systems and related data security obligations.

To notify OpenAI that you believe this job posting is non-compliant, please submit a report through this form. No response will be provided to inquiries unrelated to job posting compliance.

We are committed to providing reasonable accommodations to applicants with disabilities, and requests can be made via this link.

OpenAI Global Applicant Privacy Policy

At OpenAI, we believe artificial intelligence has the potential to help people solve immense global challenges, and we want the upside of AI to be widely shared. Join us in shaping the future of technology.