How We Can Use AI in Teaching and Research

Lunch and Learn — Session 2

Kerry Back, Rice University

The AI Agent Landscape

OpenAI Codex

• Terminal-based coding agent
• Full local file access
• Runs code, edits files, uses tools
• Free with OpenAI account

Google Gemini CLI

• Terminal-based agent
• Google ecosystem integration
• Code execution, file access
• Free tier available

Anthropic Claude Code

• Terminal-based universal agent
• Full local file access + internet
• Skills, MCP connectors, subagents
• Also available as Claude Desktop (Code tab)

All three are AI agents with tools — not chatbots. This session focuses on Claude Code and Claude Desktop, but the concepts (skills, connectors, iteration) transfer to any agent platform.

I want to start by acknowledging the landscape. OpenAI, Google, and Anthropic all offer AI agents with tools — not just chatbots. I’ll focus on Claude Code and Claude Desktop because that’s what I use, but the concepts transfer. The key point: these are all agents, not chatbots.

Claude Desktop: Three Modes

Chat (Analysis)

• Sandboxed Python in the browser
• No local file access
• Must upload data manually
• Good for quick analysis

Cowork

• Runs in a local VM
• Files synced to/from VM
• No internet access (sandbox)
• Safe for sensitive work

Code

• Runs on your machine
• Full local file access
• Full internet access
• Same engine as Claude Code CLI

The Code tab in Claude Desktop and the terminal command claude use the same engine. Code mode gives Claude direct access to your machine — no sandbox, no VM, no upload step.

Claude Desktop has three modes. Chat is sandboxed — good for quick questions. Cowork runs in a VM with no internet — safe for sensitive data. Code runs directly on your machine with full access. The Code tab and the terminal command claude are the same engine.

Claude Code: Universal Agent

What makes it universal

• Full access to your local file system, internet, and installed tools
• Add instructions via CLAUDE.md — Claude reads this on every startup
• Add capabilities via skills — text files Claude reads only when needed
• Add external tools via MCP connectors — email, calendar, databases, browsers
• Run subagents in parallel for complex tasks
• One interface for everything

Claude Code is a universal agent. You extend it with text files (CLAUDE.md, skills) and connectors (MCP). The result is a single conversational interface for your entire workflow — slides, email, grading, research, file management.

The Iteration Principle

Don’t review AI’s first draft. Ask it to critique itself and revise. Repeat until stable. Then review the final version. You review a final draft instead of a first draft.

Vague — Weak

• “Make it better”
• “Improve the analysis”
• “Make it more professional”

Specific — Strong

• “Lead with the biggest risk to Q4 revenue”
• “Add year-over-year comparisons for each metric”
• “Cut the background section and start with the recommendation”

The AI’s time is cheap — a few cents per revision cycle. Your time is expensive. Let the AI iterate on itself before you invest your attention. Specific feedback produces specific improvement; vague feedback produces nothing.

This is the single most important technique for working with any AI agent. Don’t review the first draft — that’s the worst version. Ask the AI to critique itself, revise, and repeat. Then review the final version. And when you give feedback, be specific.

Slide Creation

Every faculty member spends enormous time on slides. Let me show you three approaches, all driven by Claude Code.

Slide Creation Workflows

Beamer (LaTeX)

• AI writes the .tex file
• Compiles to PDF
• Generates PNG of each slide
• Reviews for formatting issues (overflow, spacing)
• Edits and repeats autonomously
• Create a skill with your style preferences

PowerPoint

• AI generates .pptx via Python (python-pptx)
• Full control over layouts, charts, formatting
• Works from verbal descriptions
• Good for institutional templates
• Can read and modify existing decks

Quarto → reveal.js

• Markdown-based — simpler than LaTeX, more versatile
• Produces HTML slides, websites, books
• Accepts LaTeX math notation
• Convert to PDF via decktape for annotation
• AI views slides in browser and self-reviews

Three paths. Beamer if you’re in the LaTeX ecosystem. PowerPoint if you need institutional templates. Or Quarto, which is what I now use — simpler than LaTeX, more versatile, and Claude can view the rendered slides in a browser to self-review.

Demo: Autonomous Slide Creation

LIVE DEMO

I’ll ask Claude Code to:

1. Create a Beamer slide deck on a topic
2. Compile it to PDF
3. Generate PNG images of each slide
4. Review each image for formatting issues (overflow, spacing, alignment)
5. Fix any problems and recompile
6. Repeat until satisfied

The entire process runs autonomously — I review the final version.

LIVE DEMO: Ask Claude Code to create a Beamer deck, compile, generate PNGs, review, fix, and repeat. Show the autonomous loop in action. The key point: you review the final version, not every intermediate draft.

Quarto for Slides

Why Quarto

• Markdown language — simpler than TeX, similar in spirit
• Produces reveal.js HTML slides (like this deck)
• Also produces websites, online books, can run embedded code
• Claude recommended Quarto over Beamer for visual quality
• Accepts LaTeX math: $E = mc^2$

The Workflow

• Add an MCP browser connector so Claude can view the rendered slides
• Claude writes Quarto → renders → views in browser → edits → repeats
• Hypothesis for student annotation (free accounts, works like Acrobat)
• Convert to PDF via decktape if you need a file students can download
• These slides are a Quarto deck that Claude built

I now use Quarto for all my slides. Claude recommended it over Beamer for visual quality. With an MCP browser connector, Claude can view the rendered HTML slides and self-review — just like with Beamer PNGs but more interactive. Hypothesis solves the annotation problem for HTML slides.

CLAUDE.md and Skills

Let me explain the system that makes Claude Code so powerful — CLAUDE.md files and skills. These are just text files, but they transform a generic AI into your personalized assistant.

CLAUDE.md

The file Claude always reads on startup

• A text file in your project folder (or global ~/.claude/CLAUDE.md)
• Contains your preferences, project context, and rules
• Claude follows these instructions automatically — no need to repeat them every session
• Example contents: “Always use Beamer with the metropolis theme,” “My fiscal year starts in February,” “When generating charts, use blue and amber colors”

Think of CLAUDE.md as standing instructions for a junior analyst. You write them once, and Claude follows them in every conversation. You can also describe your skills here so Claude knows when to use them automatically.

CLAUDE.md is the file Claude reads on every startup. Put your preferences, project context, and rules here. Claude follows these instructions automatically in every conversation. You write them once and never repeat them.

Skills

What They Are

• Text files with reusable instructions
• Claude reads them only when needed — don’t clog the context window
• Invoke with /skill_name (slash command)
• Or describe the skill in CLAUDE.md so Claude uses it automatically
• Claude can create skills for you: “Create a skill for building Beamer decks”

What They Can Do

• Format text (status reports, summaries)
• Process data (read files, run calculations, generate charts)
• Orchestrate workflows (multiple steps, multiple files)
• Spawn subagents for parallel work
• Anything you can describe in English

Skills are just text files. The key feature: Claude reads them only when needed, so they don’t use up context. You invoke them with a slash command or let Claude decide. And Claude can create skills for you — just describe what you want.

Example: A Beamer Skill

beamer-create skill

Rules: - Use the metropolis theme with my custom color scheme - No more than 5 seconds to read any slide - No more than 3 bullet points per slide - Use section divider slides between major topics - After creating the deck, compile to PDF - Generate PNG of each slide and review for formatting - Fix any overflow, spacing, or alignment issues - Repeat compile → review → fix until no issues remain

Ask Claude to create this skill: “Create a skill called beamer-create with my Beamer preferences.” Split into beamer-create and beamer-review if you don’t always need the review loop. Every deck you produce will be consistent.

Here’s what a Beamer skill looks like. Rules for formatting, plus instructions to compile, generate PNGs, review, and fix autonomously. Ask Claude to create it for you. Every deck you produce will follow the same rules.

The Critique Skill

Reviewer 1: Correctness

• Factual errors?
• Logical gaps?
• Missing information?
• Claims supported?

Reviewer 2: Clarity

• Logical structure?
• Anything confusing or buried?
• Main message direct enough?
• Redundancy?

Reviewer 3: Devil’s Advocate

• Strongest counterarguments?
• Weakest reasoning?
• What would a skeptic challenge?
• Alternative interpretations?

Invoke with /critique filename. Spawns three subagents in parallel — each reviews from a different angle, then synthesizes findings and applies revisions. Run in a loop: critique → fix → critique again until stable. Works on papers, slide decks, grant proposals, course materials.

The critique skill is one of the most useful things I’ve built. Three parallel reviewers — correctness, clarity, and devil’s advocate. Invoke it on any file. It runs in a loop, critiquing and fixing until no further improvements are needed. I use it on all my papers and slide decks.

Subagents

What They Are

• Claude spawns independent agents to work on subtasks
• They run in the background — you continue your conversation
• Claude decides when to use them spontaneously (parallelizable tasks)
• Or you can direct: “Use subagents to research these three topics in parallel”

Examples

• Critique skill: three reviewers run simultaneously
• Research: search for papers on three topics at once
• Code: test multiple approaches in parallel
• File processing: convert 20 documents simultaneously
• Each subagent has its own context — doesn’t clog the main conversation

Subagents are how Claude parallelizes work. The critique skill uses three subagents. But Claude also spawns them spontaneously when it recognizes a task that benefits from parallelization. You continue your conversation while they work.

MCP Connectors

MCP — Model Context Protocol — is how you connect Claude to external services. Let me show you the most useful ones for faculty.

Email & Calendar

Gmail MCP Connector

• Reads both email accounts (Rice institutional + personal)
• Manages both calendars — creates events, checks conflicts
• Drafts replies, identifies emails needing responses
• Claude walks you through the one-time setup

“Are there any events in my emails that should be added to my calendars? If so, add them.”

“Are there any emails that I need to reply to?”

“Draft an email to [person] about [topic] saying [key points].”

Rice’s security policy makes merging email accounts difficult, but it doesn’t restrict API access. Claude can read and manage both accounts from a single conversational interface.

I have two Gmail accounts — Rice and personal — and Rice’s security policy makes merging them difficult. But the security policy doesn’t restrict API access. Claude reads both, manages both calendars, drafts replies. One interface instead of switching between accounts.

Canvas API

Setup

• Request an API key once (Claude tells you how to find the link in Canvas)
• After that: no more SSO login, no more two-factor authentication
• Claude handles all Canvas interactions via the API

What Claude Can Do

• Upload files and assignments
• Download student submissions
• Upload gradebook and comments
• Help with grading — provide a rubric and let Claude assess
• Ask for a summary of why Claude assigned each grade

FERPA compliance: Anonymize submissions before grading. Ask for a script to upload a dummy gradebook first, then run the script on the actual gradebook. Student identity is stripped before AI processing.

Canvas has an API. You request a key once, and from then on Claude handles all the file management. For grading, anonymize first, grade with AI assistance, then de-anonymize. Test with a dummy gradebook before running on real grades.

Git & File Search

Version Control

• Ask Claude to initialize a git repo and commit periodically
• Each commit is a snapshot of your work
• Search past versions: “I used to have a paragraph that discussed … Find it”
• Much easier than searching old Dropbox or Overleaf versions
• Push to GitHub for backup, sharing, or websites (GitHub Pages)

File Search

• “I have a docx somewhere in my Dropbox folder that discusses … Find it”
• Claude searches your entire machine with vague clues
• Works across all file types
• Git history search: find deleted paragraphs, old versions of analyses
• You don’t need to know git — Claude handles the commands

Git is great but complex. Fortunately, Claude knows all about it, so you don’t need to know anything. Ask Claude to initialize a repo, commit periodically, and search past versions. File search across your entire machine with vague clues is also extremely useful.

Task Management

One fewer app to manage

• Ask Claude to create a task skill with categories: personal, teaching, research
• “What are my open teaching tasks?”
• “Add a task to review the midterm exam by Friday”
• “What’s overdue?”
• Stored as files — Claude reads and updates them via the skill

There are many to-do apps, but it simplifies things to do everything inside Claude Code. The way all of personal computing will soon work, I think.

You can keep your task list inside Claude Code. Create a task skill with categories, then manage everything from conversation. One fewer app to switch between.

VS Code

If you’re going to adopt one new tool beyond Claude Code itself, VS Code is the one I’d recommend.

VS Code: One App for Everything

VS Code + Claude Code extension

• Code: Python, R, LaTeX, Quarto — all in one editor
• Generate: Figures, tables, charts — AI writes the code and runs it
• Compose: “Generate a figure showing X and insert it into the LaTeX file at section 3”
• Orchestrate: Claude can edit multiple files, compile, review output, and iterate
• Remote: SSH into a server and run Claude Code there via VS Code

Claude can generate a matplotlib figure, save it, insert the \includegraphics command into your .tex file, compile the PDF, and review the result — all from a single instruction. One application for your entire research and teaching workflow.

VS Code with the Claude Code extension turns one application into a complete workflow. You can say “generate a figure showing the efficient frontier and insert it into my slide deck” and Claude will write the Python, generate the figure, edit the LaTeX file, compile, and review. It orchestrates across files and tools.

VS Code + Remote Server (jgsrc1)

Setup

• Install the Remote SSH and Claude Code extensions in VS Code
• SSH into jgsrc1 from within VS Code
• Install Claude Code on the server: curl -fsSL https://claude.ai/install.sh | bash
• Run claude in the VS Code terminal

What This Gets You

• Edit files on the server with AI assistance
• Run computations on jgsrc1’s hardware
• Claude Code operates on the server’s file system directly
• VS Code runs on your laptop; everything else runs remotely
• Same experience as local — just on a more powerful machine

If you do computation-heavy work, run Claude Code on the school server. Install Remote SSH in VS Code, connect to jgsrc1, install Claude Code there. Full AI agent on server hardware while you interact from your laptop.

AI for Research

Now let me shift to the research side. AI agents are changing research workflows in several concrete ways.

Research Workflows

Literature & Documents

• Upload papers to NotebookLM — query across 50 sources with citations
• Audio Overview generates a podcast-style discussion of your research
• Compare year-over-year changes in 10-K disclosures or policy documents
• AI reads and summarizes papers, identifies methodology gaps
• RAG pipelines for large document collections

Empirical Analysis

• AI writes Python/R code for econometric and statistical analysis
• Generates figures and tables, inserts them into your manuscript
• Data cleaning, merging, and wrangling from verbal descriptions
• Replication and robustness checks directed through conversation
• The critique skill catches weaknesses before reviewers do

Two broad categories. Literature and document analysis: NotebookLM lets you upload 50 papers and query with inline citations. Empirical work: the AI writes your analysis code, generates figures and tables, and can insert them directly into your LaTeX manuscript.

The Research Loop in Practice

📄 Read AI summarizes papers, identifies gaps and methods

💻 Analyze AI writes code, generates figures and tables

📝 Write Draft sections, insert results into LaTeX

🔍 Critique Three-reviewer critique skill finds weaknesses

Each cycle takes minutes, not days. The AI remembers context across the session, so you can iterate rapidly. The critique skill is particularly valuable — three parallel reviewers catch issues you’d miss on your own.

The full research loop. Read and summarize the literature. Write analysis code, generate results. Draft the paper, insert figures and tables. Then run the critique skill — three parallel reviewers. Each cycle takes minutes.

Demo: Research Workflow

LIVE DEMO

I’ll show Claude Code:

1. Reading a dataset and exploring its structure
2. Running a regression analysis
3. Generating a formatted results table
4. Creating a figure (e.g., coefficient plot)
5. Inserting both the table and figure into a LaTeX paper
6. Compiling the PDF and reviewing the output

LIVE DEMO: Show the full research loop — from raw data to figures and tables inserted into a LaTeX paper. The key point: Claude orchestrates across Python (for analysis), LaTeX (for the paper), and the file system (for saving outputs).

Rethinking Assessment

We’ve seen that AI can produce deliverables, manage workflows, and support research. What does this mean for how we assess students?

The Problem

Current State

• Most assessments test production
• Write a report, build a model, create a deck
• AI does all of this in 60 seconds
• We are testing a skill that has been commoditized

The Question

• “Did the student do the work?” is now the wrong question
• The right question: can they defend it?
• Can they explain assumptions?
• Can they field hard questions?
• Can they catch what AI got wrong?

This applies most directly to analytical and data-intensive courses. In disciplines where producing the work IS the learning, the framework needs adaptation.

When AI can produce a professional deliverable in under a minute, what are we actually testing? The durable skill is verification and defense.

The AI Examiner

📄 Upload Student submits slides (PDF)

🔍 Pre-Analysis AI identifies key claims, gaps, question areas

🎙️ Voice Exam 10–12 adaptive questions via voice AI

📊 Grading Council Three AI models grade independently, then deliberate

Multi-model deliberation

• Three frontier models grade independently, see peers’ scores, adjust with reasoning
• Faculty review required for borderline grades and all appeals
• Convergence metrics are auditable
• Practice mode available 24/7

The flow: upload slides, AI pre-analyzes, voice AI conducts adaptive questioning, three models grade and deliberate. Faculty review all borderline grades.

The Presentation Examiner

How It Works

• Students receive a magic-link login (no passwords)
• Upload PDF slides
• Session 1: Present to AI listener (ElevenLabs voice agent)
• Session 2: Answer 3 AI-generated questions + follow-ups
• Claude Sonnet analyzes slides and generates questions
• GPT-4 grades on three rubrics with detailed feedback

What It Solves

• Traditional: schedule every student for a live presentation (nightmare)
• This system: anytime, anywhere presentations
• Instant, consistent grading with detailed rubric feedback
• Instructor reviews transcripts and grades later
• Scales from 20 to 2,000 students without hiring graders
• Live now — built for the Rice executive education program

This is our implementation. Students get a magic-link login, upload slides, and present to an AI voice agent. Claude generates questions from the slide content, GPT-4 grades on three rubrics. The instructor reviews transcripts and grades. It scales without hiring graders.

Concerns and Validation

Cost

• API costs: ~$1 per exam at current pricing
• Development and maintenance: ongoing
• Faculty review time for borderline grades
• Total cost of ownership exceeds API cost
• But scales better than human grading

Fairness

• Practice mode available 24/7
• Extended time accommodations built in
• Text-based alternative available
• Accent/fluency bias testing: explicit pilot deliverable
• FERPA review planned

Validation Plan

• Blind AI vs. faculty grading comparison
• Pre-registered reliability threshold (Cohen’s kappa)
• No deployment without pilot evidence
• Student feedback and learning outcomes
• Target: 2027–28, pending results

Transparent about costs: total cost of ownership exceeds the API cost. Bias testing is an explicit deliverable, not an afterthought. Validation is rigorous: blind comparison, pre-registered thresholds, no deployment without evidence.

Who Wants to Pilot This?

I’m looking for 2-3 faculty across different disciplines to pilot this in Spring 2027. You keep full control of your course. We add the AI examiner as one assessment component and compare results. Who’s interested?