agentic AI in EEB
BIOEE 7600-103 · Fall 2026 · a graduate seminar at Cornell
Live page — updated through the semester
An agent is a large language model given tools, memory, and permission to plan and act over many steps — cleaning datasets, running analyses, writing and executing code, querying databases, monitoring the literature, drafting the outputs. That autonomy is what makes agents useful, and what raises the stakes when they go wrong.
This seminar is a hands-on and deliberately skeptical tour of what agents can and can’t be trusted to do across the research lifecycle. After a week defining what “agentic” actually means, most of each session is a live demo: we give an agent a real EEB task and then audit the process and the result, hunting for what it got wrong or quietly hid. We close by drafting an EEB community guideline for responsible use.
No computer-science background is assumed. A little R or Python helps you follow the coding demos but isn’t required. Bring a laptop.
Start with what is an agent? — a ten-minute primer written for this seminar, covering the vocabulary and the core ideas with no background assumed. It is the first of the Week 1 readings; read it before the rest.
This week
What agentic AI is (and isn't) — from LLMs to tool-using agents
Framing
What actually changes when an AI can use tools and take actions instead of only producing text — and how is that different from an LLM, from deep learning, and from a fixed pipeline?
Read before class
- What is an agent? A primer for this seminar — Benning & Xu — start herefoundationsWritten for this course, for people with no background in any of this. Ten minutes: the ladder from a plain language model up to an agent, the loop every agent runs, why read-only and read-write agents carry categorically different risk, and the twenty terms you will hear all term. Read this one first.
- A new Moore's Law for AI agents — AI DigestfoundationsevaluationFive minutes. How long a task an agent can finish on its own, plotted over time. Then check its headline number against METR's own January re-fit — the popular version of this chart is already stale, and noticing that is the exercise.
- The Dark Night of Mathematics — Kirwin Hampshire, July 2026debateA pure-mathematics PhD student on what it feels like when the machine gets there first. Mathematics is the field furthest down this road; read it as a report from somewhere we may be going.
- A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis — Xu, Ren, Wang et al. 2025, Nature Medicine — abstract and figuresapplicationThe strongest positive case on the table: an AI-found target and an AI-designed molecule, a dose-dependent lung-function signal in 71 patients, and a phase III trial opened in July 2026. Note also whose drug it is and who ran the trial. Shorter way in — the authors' "Behind the Paper" post.
- Machines of Loving Grace — §1, Biology and health — Dario Amodei, October 2024 — section 1 onlydebateThe optimist case at full strength: 50 to 100 years of biological progress compressed into 5 to 10. Read the section opening and the list of expected breakthroughs; skim the rest.
- No, Dario Amodei, we will not be curing cancer and most human disease — Gary Marcus, August 2026debateThe rebuttal, three weeks old, with Topol, Pachter, Robison and others on the record — and the argument that clinical timelines, not intelligence, are the binding constraint. Read against the trial above and decide who is right.
Bring with you
Come with one question you would actually want an agent to answer from your own research. We will try a few of them live.
Schedule
Nine themed weeks. Dates firm up as the term does; expand a week for its framing question, readings, and demo.
Week 1 What agentic AI is (and isn't) — from LLMs to tool-using agents
Foundations · September 4, 2026
What actually changes when an AI can use tools and take actions instead of only producing text — and how is that different from an LLM, from deep learning, and from a fixed pipeline?
Demo
A chatbot becomes an agent, live: first watch a model generate plausible text rather than verified fact, then give an agent a goal that forces it to reach into the world and pull real occurrence records.
Readings
- What is an agent? A primer for this seminar — Benning & Xu — start herefoundationsWritten for this course, for people with no background in any of this. Ten minutes: the ladder from a plain language model up to an agent, the loop every agent runs, why read-only and read-write agents carry categorically different risk, and the twenty terms you will hear all term. Read this one first.
- A new Moore's Law for AI agents — AI DigestfoundationsevaluationFive minutes. How long a task an agent can finish on its own, plotted over time. Then check its headline number against METR's own January re-fit — the popular version of this chart is already stale, and noticing that is the exercise.
- The Dark Night of Mathematics — Kirwin Hampshire, July 2026debateA pure-mathematics PhD student on what it feels like when the machine gets there first. Mathematics is the field furthest down this road; read it as a report from somewhere we may be going.
- A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis — Xu, Ren, Wang et al. 2025, Nature Medicine — abstract and figuresapplicationThe strongest positive case on the table: an AI-found target and an AI-designed molecule, a dose-dependent lung-function signal in 71 patients, and a phase III trial opened in July 2026. Note also whose drug it is and who ran the trial. Shorter way in — the authors' "Behind the Paper" post.
- Machines of Loving Grace — §1, Biology and health — Dario Amodei, October 2024 — section 1 onlydebateThe optimist case at full strength: 50 to 100 years of biological progress compressed into 5 to 10. Read the section opening and the list of expected breakthroughs; skim the rest.
- No, Dario Amodei, we will not be curing cancer and most human disease — Gary Marcus, August 2026debateThe rebuttal, three weeks old, with Topol, Pachter, Robison and others on the record — and the argument that clinical timelines, not intelligence, are the binding constraint. Read against the trial above and decide who is right.
Bring with you
Come with one question you would actually want an agent to answer from your own research. We will try a few of them live.
Week 2 Agentic coding & data analysis
The research data lifecycle · September 11, 2026
The agent wrote the code and found a significant effect. Should you believe either — especially if you told it what you were hoping to find?
Readings
- A case study of evaluating AI agents on a neuroscience data-to-discovery pipeline — Horstmann, Lin, Robie, Sun & Branson 2026, COLM (arXiv:2606.07718)evaluationapplicationThe closest thing yet to this course, done as research. Coding agents turned loose on a real fly-optogenetics pipeline and scored stage by stage against the criteria the scientists who built it actually use. Read the abstract, Figure 2 for the seven stages and what counted as passing each, Table 1 for who passed what, then section 4.2. Two stages were solved by every agent; body tracking was failed by every agent in every trial; and stages solved in isolation fell over once composed end to end. Come ready to argue about the visualization audit — the authors classified every episode in which an agent looked at a plot of its own output, and failures outnumbered fixes on every single task. Run on Opus 4.6 and GPT-5.4, so "not yet" here is not a stale result about last year's models.
Week 3 Agentic data exploration, compilation & archiving
The research data lifecycle · September 18, 2026
An agent can compile and standardize a messy dataset in seconds. How do you know it didn't quietly break it — and can anyone, including future you, reconstruct what it did?
Week 4 Agentic literature synthesis
Literature synthesis & writing · September 25, 2026
Where does an agent legitimately speed up finding, screening, and synthesizing literature — and where does it quietly corrupt the record?
Week 5 Agentic scientific writing
Literature synthesis & writing · October 2, 2026
Where does AI-assisted drafting genuinely help — including for researchers writing in a second language — and where does it cross into misrepresentation?
Week 6 Project & lab management and automation
Lab management, teaching & training · October 9, 2026
Which lab and project tasks should we hand to an agent — and what has to stay under direct human control?
Week 7 Teaching, mentoring & implications for graduate training
Lab management, teaching & training · October 16, 2026
If a student can generate the answer in seconds, what are we actually teaching — and does leaning on agents accelerate skill-building or quietly prevent it?
Week 8 Costs, ethics & accountability of agentic AI
Costs, ethics & governance · October 23, 2026
Agents make many model calls, not one — so as ecologists, how should their environmental and ethical footprint shape our use of them? And when an autonomous system errs, who is accountable?
Week 9 Formulating an EEB community guideline
Costs, ethics & governance · October 30, 2026
What should our department collectively expect of anyone using agents in research, writing, teaching, and lab work?
Reading room
Everything worth reading, assigned or not, added as we find it. Assigned items carry the week they belong to; filter by topic to find the rest.
- What is an agent? A primer for this seminarweek 1foundations
Benning & Xu — start here Written for this course, for people with no background in any of this. Ten minutes: the ladder from a plain language model up to an agent, the loop every agent runs, why read-only and read-write agents carry categorically different risk, and the twenty terms you will hear all term. Read this one first. - A new Moore's Law for AI agentsweek 1foundationsevaluation
AI Digest Five minutes. How long a task an agent can finish on its own, plotted over time. Then check its headline number against METR's own January re-fit — the popular version of this chart is already stale, and noticing that is the exercise. - The Dark Night of Mathematicsweek 1debate
Kirwin Hampshire, July 2026 A pure-mathematics PhD student on what it feels like when the machine gets there first. Mathematics is the field furthest down this road; read it as a report from somewhere we may be going. - A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosisweek 1application
Xu, Ren, Wang et al. 2025, Nature Medicine — abstract and figures The strongest positive case on the table: an AI-found target and an AI-designed molecule, a dose-dependent lung-function signal in 71 patients, and a phase III trial opened in July 2026. Note also whose drug it is and who ran the trial. Shorter way in — the authors' "Behind the Paper" post. - Machines of Loving Grace — §1, Biology and healthweek 1debate
Dario Amodei, October 2024 — section 1 only The optimist case at full strength: 50 to 100 years of biological progress compressed into 5 to 10. Read the section opening and the list of expected breakthroughs; skim the rest. - No, Dario Amodei, we will not be curing cancer and most human diseaseweek 1debate
Gary Marcus, August 2026 The rebuttal, three weeks old, with Topol, Pachter, Robison and others on the record — and the argument that clinical timelines, not intelligence, are the binding constraint. Read against the trial above and decide who is right. - A case study of evaluating AI agents on a neuroscience data-to-discovery pipelineweek 2evaluationapplication
Horstmann, Lin, Robie, Sun & Branson 2026, COLM (arXiv:2606.07718) The closest thing yet to this course, done as research. Coding agents turned loose on a real fly-optogenetics pipeline and scored stage by stage against the criteria the scientists who built it actually use. Read the abstract, Figure 2 for the seven stages and what counted as passing each, Table 1 for who passed what, then section 4.2. Two stages were solved by every agent; body tracking was failed by every agent in every trial; and stages solved in isolation fell over once composed end to end. Come ready to argue about the visualization audit — the authors classified every episode in which an agent looked at a plot of its own output, and failures outnumbered fixes on every single task. Run on Opus 4.6 and GPT-5.4, so "not yet" here is not a stale result about last year's models. - Building effective agentsfoundations
Anthropic, December 2024 Optional. The engineering account of what an agent is and the case for not building one — the workflow-versus-agent distinction is the part worth having. From December 2024, so it predates most of what we will actually use, and it is written for people building these systems rather than using them. Worth a skim once the primer has done its job. - The Hugging Face hack is a PR crisis that's costing OpenAI millionssecurity
Fortune, August 2026 An agent given a goal routed around the barriers meant to contain it, and four other services turned out to have been breached too. No attacker anywhere in the story — the agent was optimizing exactly as instructed. Our Week 8 case. - The Hugging Face incident exposes the wrong AI security debatesecuritydebate
Shlomo Kramer in Fortune, August 2026 The counterpoint: a human insider threat unfolds over weeks and leaves a pattern; an agent takes thousands of actions before anyone notices. - Emerging applications of large language models in ecology and conservation scienceEEBapplicationethics
Mammides et al. 2026, Conservation Biology (open access) Our own literature, finally. A review by conservation scientists covering data extraction, evidence synthesis, monitoring and coding, with a long and honest section on hallucination, bias, reproducibility, equity, deskilling and energy cost. Mostly about LLMs rather than agents — which is itself the state of the field. - Large language model agents for biological intelligenceapplicationevaluation
Dip et al. 2026, Briefings in Bioinformatics (open access) A survey that takes agentic behaviour itself as the unit of analysis across genomics, proteomics, spatial biology and biomedicine. Its verdict is the useful part: brittle tool interfaces, silent error propagation, poor reproducibility, and not mature enough for autonomous deployment. - The End of Mathematicsdebate
Daniel Litt, August 2026 The institutional version of Hampshire's essay: a field that keeps rewarding papers while the machines write them, stops training anyone, and quietly dies. - The AI revolution in math has arriveddebate
Quanta Magazine, April 2026 Free to read, and calibrated: real verified results alongside Tao on "scattered successes among a big sea of unreported failures." - Agentic AI, explainedfoundations
MIT Sloan Ideas Made to Matter A plain-language explainer if the Anthropic piece assumes too much. - Mathematicians are developing rules for AI use — other fields should followpolicydebate
Nature news A worked example of another discipline doing exactly what we will attempt in Week 9. - Ten simple rules for using large language models in science, version 1.0EEBpolicy
Smith et al. 2024, PLOS Computational Biology Written by ecologists. The closest thing we have to a starting point for our own guideline. - Fabrication and errors in the bibliographic citations generated by ChatGPTevaluationreproducibility
Walters & Wilder 2023, Scientific Reports The evidence behind the course rule that an agent may draft, but only a human may cite. - Generative AI's environmental costs are soaring — and mostly secretethics
Crawford 2024, Nature - The FAIR Guiding Principles for scientific data management and stewardshipreproducibility
Wilkinson et al. 2016, Scientific Data - Deep learning as a tool for ecology and evolutionEEBfoundations
Borowiec et al. 2022, Methods in Ecology and Evolution Background on the domain tools an agent would orchestrate — the substrate, not the agent. - Cornell AI guidelinespolicy
IT@Cornell The data rules we follow in this course. Read the part about what you may not paste into an external tool.
Nothing tagged that yet.
How each session works
A discussion seminar built around live demos, not a lecture course. Roughly: five minutes on the framing question, fifteen on the reading, thirty-five on a live agent demo and the audit that follows it, five to wrap up.
Audit of the Week. In rotation, each of us brings one real case of an agent doing something wrong in our own subfield — a bad clean, a spurious result, a fabricated reference, a silent overwrite. Two minutes, then on with the demo. Skepticism as muscle memory. The cases we accumulate become raw material for the guideline we draft in Week 9.
Using AI in this course
We practice what we study. You are encouraged to use agentic tools throughout, and expected to be transparent: name the tool and version, and say what it did. You remain fully accountable for accuracy, including every citation and every number an agent produced. That is Cornell’s guiding principle — you are accountable for your work regardless of the tools you use — along with Cornell’s data rule: never paste unpublished data, personal information, reviewer identities, or institutional data into free, non-Cornell AI tools.
Because agents act, we add one rule of our own: gate every irreversible action — delete, overwrite, submit, send — behind human confirmation.
Getting a seat
Enrollment is capped at 20 and by instructor consent; undergraduates are not eligible. Email Xiangtao or John and we will add you. Postdocs and faculty are welcome to join the discussion without registering.