Position Description

Postdoctoral Fellows — Agents that Read, Reason, and Act

Reading-to-Learn (R2L) Lab, Cheriton School of Computer Science, University of Waterloo

Supervisor: Victor Zhong

Start date: Flexible, ideally Fall 2026; rolling review until filled

Term: 2-year initial term, with a 3rd year normally extended subject to satisfactory progress and continued funding

Location: Waterloo, Ontario, Canada (in person)

The opportunity

The R2L Lab at the University of Waterloo is hiring postdoctoral fellows to lead research at the frontier of agentic AI: systems that read, reason, and act on knowledge to accomplish open-ended tasks.

We are looking for researchers who will define a research agenda, not execute one. Successful applicants will lead a small team of PhD and MMath students, set technical direction, and drive projects toward publication at top venues (NeurIPS, ICML, ICLR, TMLR, ACL, EMNLP, COLM). The lab provides substantial compute, open-benchmark and open-source commitments, access to applied deployment partners, and direct mentorship from the PI.

What you will work on

We expect each fellow to own a research thread that advances one or more of the following capability areas. The list is directional, not prescriptive: strong candidates with adjacent agendas are encouraged to apply and to propose their own framing.

• Trustworthy and governable agents. Today's agents are difficult to inspect, govern, or constrain. We are interested in the foundations that would make an agent's capabilities and access patterns into first-class, reviewable artifacts rather than implicit behavior. Questions span how agents represent what they can and cannot do, how humans engage with those representations, and what kinds of guarantees compose across systems and across time.
• Adaptation without retraining. The environments in which agents operate change faster than models can be retrained. We are interested in how agents can take on new contexts — new data, new tools, new constraints — with little or no supervision. Questions span what an agent should absorb versus retrieve, how adaptation interacts with safety and reliability, and what theory has to say about learning at deployment time.
• Information access for agents. Retrieval, memory, and evidence systems were designed for human users; agents are a different kind of consumer. We are interested in how these systems should be reconceived when the user is an autonomous reasoner rather than a person. Questions span what the right unit of information is, what signals should drive access, and how to evaluate these systems when downstream task success is the metric that matters.
• Computer-use agents. Many real tasks live behind interfaces that no API exposes. We are interested in agents that perceive and operate general software with the reliability and breadth of a skilled human user. Questions span perception, control, robustness, long-horizon behavior, and how agents acquire and retain skills from realistic operating environments.

These threads are method-first. Applied partners give us realistic data, workflows, and deployment surfaces, but the science is domain-general and the artifacts are open.

Who we are looking for

Required.

• PhD (in hand or imminent) in computer science, statistics, computational science, or a closely related field.
• A track record of first-author publications at top-tier ML/NLP venues. Applicants without first-author publications at NeurIPS, ICML, ICLR, TMLR, ACL, EMNLP, COLM, CORL, RSS, CVPR, ICCV, MLSys, OSDI will not be considered.
• Demonstrated ability to take a project from idea to artifact independently: problem formulation, experiment design, implementation, writing, and rebuttal.
• Strong engineering practice. You can prototype, profile, and ship.

Strongly preferred.

• Evidence of research leadership: mentoring junior students, leading multi-author papers, organizing workshops, or driving open-source releases.
• Experience with one or more of: large language models, retrieval systems, reinforcement learning, multimodal models, agent benchmarks, or human evaluation.
• A clear, ambitious vision for the next 3–5 years of your research program.

Not required.

• Prior domain expertise in any specific application area. We will connect you with the right collaborators.
• A 1-to-1 match with the capability areas above. Tell us what you would do.

What we offer

• Compensation. CAD $80,000–$100,000 per year, commensurate with experience, plus full University of Waterloo benefits.
• Compute. Substantial GPU allocation through lab clusters, the Vector Institute, and partner infrastructure.
• Mentorship. Weekly 1:1s with the PI, an engaged cohort of 5+ senior PhDs, and access to collaborators across academia and industry.
• Leadership runway. You will co-mentor PhD and MMath students, co-author with the PI as a peer, and have a direct path to faculty applications, industry research labs, or founding roles. We actively support the job market.
• Travel. Funded conference travel and collaboration trips.
• Translation. Active partnerships with industrial deployment teams give your work a realistic deployment surface when you want one.

How to apply

Apply via AcademicJobsOnline with the following materials:

1. CV. Include publications, software, talks, mentees, and service.
2. Research statement (max 4 pages). Tell us what you would lead. Past, present, and proposed work; we weight the proposed section heavily and care more about ambition and feasibility than about coverage of the capability areas above.
3. Two representative publications. First-author preferred. Preprints are fine.
4. Two recommendation letters. From research advisors and collaborators.

Applications are reviewed on a rolling basis until all positions are filled.

Questions: victor.zhong@uwaterloo.ca.

About the lab

The Reading-to-Learn (R2L) Lab studies how agents read, reason, and act on knowledge. Recent work spans retrieval, compositional reasoning, computer-use agents, robustness, and evaluation. Our publications appear at NeurIPS, ICML, ICLR, TMLR, ACL, EMNLP, and COLM. We work closely with the broader Ontario AI ecosystem (Vector Institute, Waterloo AI Institute) and with industry partners across the AI research community.

Equity, diversity, and inclusion

The University of Waterloo acknowledges that much of our work takes place on the traditional territory of the Neutral, Anishinaabeg, and Haudenosaunee peoples. The University of Waterloo is committed to implementing the Calls to Action framed by the Truth and Reconciliation Commission.

We encourage applications from all qualified candidates, including women, members of racialized minorities, Indigenous peoples, persons with disabilities, persons of any sexual orientation or gender identity, and others who may contribute to the further diversification of ideas. All qualified candidates are encouraged to apply.