Skip to main content
Chemistry LibreTexts

8.1: What are AI Agents?

  • Page ID
    555730
  • ​

    ​

    \( \newcommand{\vecs}[1]{\overset { \scriptstyle \rightharpoonup} {\mathbf{#1}} } \)

    \( \newcommand{\vecd}[1]{\overset{-\!-\!\rightharpoonup}{\vphantom{a}\smash {#1}}} \)

    \( \newcommand{\dsum}{\displaystyle\sum\limits} \)

    \( \newcommand{\dint}{\displaystyle\int\limits} \)

    \( \newcommand{\dlim}{\displaystyle\lim\limits} \)

    \( \newcommand{\id}{\mathrm{id}}\) \( \newcommand{\Span}{\mathrm{span}}\)

    ( \newcommand{\kernel}{\mathrm{null}\,}\) \( \newcommand{\range}{\mathrm{range}\,}\)

    \( \newcommand{\RealPart}{\mathrm{Re}}\) \( \newcommand{\ImaginaryPart}{\mathrm{Im}}\)

    \( \newcommand{\Argument}{\mathrm{Arg}}\) \( \newcommand{\norm}[1]{\| #1 \|}\)

    \( \newcommand{\inner}[2]{\langle #1, #2 \rangle}\)

    \( \newcommand{\Span}{\mathrm{span}}\)

    \( \newcommand{\id}{\mathrm{id}}\)

    \( \newcommand{\Span}{\mathrm{span}}\)

    \( \newcommand{\kernel}{\mathrm{null}\,}\)

    \( \newcommand{\range}{\mathrm{range}\,}\)

    \( \newcommand{\RealPart}{\mathrm{Re}}\)

    \( \newcommand{\ImaginaryPart}{\mathrm{Im}}\)

    \( \newcommand{\Argument}{\mathrm{Arg}}\)

    \( \newcommand{\norm}[1]{\| #1 \|}\)

    \( \newcommand{\inner}[2]{\langle #1, #2 \rangle}\)

    \( \newcommand{\Span}{\mathrm{span}}\) \( \newcommand{\AA}{\unicode[.8,0]{x212B}}\)

    \( \newcommand{\vectorA}[1]{\vec{#1}}      % arrow\)

    \( \newcommand{\vectorAt}[1]{\vec{\text{#1}}}      % arrow\)

    \( \newcommand{\vectorB}[1]{\overset { \scriptstyle \rightharpoonup} {\mathbf{#1}} } \)

    \( \newcommand{\vectorC}[1]{\textbf{#1}} \)

    \( \newcommand{\vectorD}[1]{\overrightarrow{#1}} \)

    \( \newcommand{\vectorDt}[1]{\overrightarrow{\text{#1}}} \)

    \( \newcommand{\vectE}[1]{\overset{-\!-\!\rightharpoonup}{\vphantom{a}\smash{\mathbf {#1}}}} \)

    \( \newcommand{\vecs}[1]{\overset { \scriptstyle \rightharpoonup} {\mathbf{#1}} } \)

    \(\newcommand{\longvect}{\overrightarrow}\)

    \( \newcommand{\vecd}[1]{\overset{-\!-\!\rightharpoonup}{\vphantom{a}\smash {#1}}} \)

    \(\newcommand{\ket}[1]{\left| #1 \right>}\)
    \(\newcommand{\bra}[1]{\left< #1 \right|}\)
    \(\newcommand{\braket}[2]{\left< #1 \vphantom{#2} \right| \left. #2 \vphantom{#1} \right>}\)
    \(\newcommand{\braopket}[3]{\left< #1 \vphantom{#2}\vphantom{#3} \right| #2 \vphantom{#1}\vphantom{#3} \left| #3 \vphantom{#1}\vphantom{#2} \right>}\)
    \(\newcommand{\qmvec}[1]{\mathbf{\vec{#1}}}\)
    \(\newcommand{\op}[1]{\hat{\mathbf{#1}}}\)
    \(\newcommand{\expect}[1]{\langle #1 \rangle}\)
    \(\newcommand{\dfn}[1]{\emph{\textbf{#1}}}\)

    \(\newcommand{\avec}{\mathbf a}\) \(\newcommand{\bvec}{\mathbf b}\) \(\newcommand{\cvec}{\mathbf c}\) \(\newcommand{\dvec}{\mathbf d}\) \(\newcommand{\dtil}{\widetilde{\mathbf d}}\) \(\newcommand{\evec}{\mathbf e}\) \(\newcommand{\fvec}{\mathbf f}\) \(\newcommand{\nvec}{\mathbf n}\) \(\newcommand{\pvec}{\mathbf p}\) \(\newcommand{\qvec}{\mathbf q}\) \(\newcommand{\svec}{\mathbf s}\) \(\newcommand{\tvec}{\mathbf t}\) \(\newcommand{\uvec}{\mathbf u}\) \(\newcommand{\vvec}{\mathbf v}\) \(\newcommand{\wvec}{\mathbf w}\) \(\newcommand{\xvec}{\mathbf x}\) \(\newcommand{\yvec}{\mathbf y}\) \(\newcommand{\zvec}{\mathbf z}\) \(\newcommand{\rvec}{\mathbf r}\) \(\newcommand{\mvec}{\mathbf m}\) \(\newcommand{\zerovec}{\mathbf 0}\) \(\newcommand{\onevec}{\mathbf 1}\) \(\newcommand{\real}{\mathbb R}\) \(\newcommand{\twovec}[2]{\left[\begin{array}{r}#1 \\ #2 \end{array}\right]}\) \(\newcommand{\ctwovec}[2]{\left[\begin{array}{c}#1 \\ #2 \end{array}\right]}\) \(\newcommand{\threevec}[3]{\left[\begin{array}{r}#1 \\ #2 \\ #3 \end{array}\right]}\) \(\newcommand{\cthreevec}[3]{\left[\begin{array}{c}#1 \\ #2 \\ #3 \end{array}\right]}\) \(\newcommand{\fourvec}[4]{\left[\begin{array}{r}#1 \\ #2 \\ #3 \\ #4 \end{array}\right]}\) \(\newcommand{\cfourvec}[4]{\left[\begin{array}{c}#1 \\ #2 \\ #3 \\ #4 \end{array}\right]}\) \(\newcommand{\fivevec}[5]{\left[\begin{array}{r}#1 \\ #2 \\ #3 \\ #4 \\ #5 \\ \end{array}\right]}\) \(\newcommand{\cfivevec}[5]{\left[\begin{array}{c}#1 \\ #2 \\ #3 \\ #4 \\ #5 \\ \end{array}\right]}\) \(\newcommand{\mattwo}[4]{\left[\begin{array}{rr}#1 \amp #2 \\ #3 \amp #4 \\ \end{array}\right]}\) \(\newcommand{\laspan}[1]{\text{Span}\{#1\}}\) \(\newcommand{\bcal}{\cal B}\) \(\newcommand{\ccal}{\cal C}\) \(\newcommand{\scal}{\cal S}\) \(\newcommand{\wcal}{\cal W}\) \(\newcommand{\ecal}{\cal E}\) \(\newcommand{\coords}[2]{\left\{#1\right\}_{#2}}\) \(\newcommand{\gray}[1]{\color{gray}{#1}}\) \(\newcommand{\lgray}[1]{\color{lightgray}{#1}}\) \(\newcommand{\rank}{\operatorname{rank}}\) \(\newcommand{\row}{\text{Row}}\) \(\newcommand{\col}{\text{Col}}\) \(\renewcommand{\row}{\text{Row}}\) \(\newcommand{\nul}{\text{Nul}}\) \(\newcommand{\var}{\text{Var}}\) \(\newcommand{\corr}{\text{corr}}\) \(\newcommand{\len}[1]{\left|#1\right|}\) \(\newcommand{\bbar}{\overline{\bvec}}\) \(\newcommand{\bhat}{\widehat{\bvec}}\) \(\newcommand{\bperp}{\bvec^\perp}\) \(\newcommand{\xhat}{\widehat{\xvec}}\) \(\newcommand{\vhat}{\widehat{\vvec}}\) \(\newcommand{\uhat}{\widehat{\uvec}}\) \(\newcommand{\what}{\widehat{\wvec}}\) \(\newcommand{\Sighat}{\widehat{\Sigma}}\) \(\newcommand{\lt}{<}\) \(\newcommand{\gt}{>}\) \(\newcommand{\amp}{&}\) \(\definecolor{fillinmathshade}{gray}{0.9}\)

    An AI coding agent is a program that reads your project files, runs terminal commands, and fixes its own mistakes without waiting for you to copy and paste. Unlike a chatbot, an agent works inside your development environment. You describe the biology question you want in plain English, and the agent writes the WeBWorK PG code, tests it against a local renderer, reads any error messages, and revises the code until the problem renders correctly.

    This style of development is sometimes called "vibe coding," a term that gained popularity in late 2025. The idea is simple: you bring the biology expertise and a clear description of the question, and the agent brings the PG syntax and debugging. You do not need to learn Perl or memorize macro names. The agent handles those details because it has access to curated documentation and a live renderer to check its work.

    Chatbots versus agents

    A chatbot (such as ChatGPT in its default web interface) generates text in a conversation window. It cannot see your files, run commands, or test whether its code actually works. An AI coding agent does all of those things. The table below summarizes the key differences.

    Capabilities that distinguish a chatbot from an AI coding agent.
    Capability Chatbot AI Agent
    Reads your project files No Yes
    Runs terminal commands No Yes
    Connects to the PG renderer No Yes
    Self-corrects when code fails No Yes
    Accesses curated WeBWorK documentation No Yes
    Tests its own output No Yes

    Why chatbots fail at WeBWorK

    Even the simplest WeBWorK question defeats a standard chatbot. In a classroom demonstration, a chatbot was asked to write a problem that asks "What is your favorite color?" with a free-response answer box. The chatbot produced code that looked reasonable but failed to render. It mixed up standard Perl syntax with PG-specific conventions and could not test or fix the result.

    The root cause is that PG (the Problem Generation language) looks like Perl but is not standard Perl. PG uses its own macros, its own document structure (DOCUMENT() through ENDDOCUMENT()), and its own markup language (PGML). A chatbot trained on general programming data treats PG as ordinary Perl and produces code that fails in subtle ways. When you ask the chatbot to fix the error, it often makes things worse because it cannot see the actual error message from the renderer.

    This creates a debugging trap: the chatbot confidently tells you the code is correct and suggests the renderer itself must be broken. You spend time troubleshooting your Docker setup, reinstalling containers, and searching forums, when the real problem was always in the generated code. For more on common errors and how to recognize them, see Common Mistakes and How to Fix Them. For renderer setup and verification, see Setting Up the PG Renderer.

    What makes agents work for WeBWorK

    An AI coding agent succeeds where a chatbot fails because it has four capabilities that close the gap between generating code and producing a working problem.

    1. Documentation access. The agent reads WeBWorK-specific reference documents instead of guessing from general Perl knowledge. These documents describe the correct macro signatures, PGML syntax, and problem structure. See Knowledge Documents for AI Agents for how to load these references.
    2. Renderer connection. The agent POSTs your PG code to a local renderer (typically running at localhost:3000) and reads the response. This gives it immediate feedback on whether the code actually works. See Connecting AI Agents to the PG Renderer for setup details.
    3. Self-correction loop. When the renderer returns an error, the agent reads the error message, identifies the problem, edits the code, and resubmits. This loop continues until the problem renders cleanly, without any action from you.
    4. File management. The agent creates, edits, and organizes .pg files directly in your project directory. You do not need to copy and paste code between a chat window and your editor.

    Available agents for WeBWorK

    Two commercial AI coding agents are currently well suited for writing WeBWorK problems. Both run as command-line tools and can connect to a local PG renderer.

    AI coding agents that support WeBWorK problem development.
    Agent Provider How it works Documentation mechanism
    Claude Code Anthropic Runs locally in your terminal Skills (curated doc bundles in .claude/skills/)
    OpenAI Codex OpenAI Cloud sandbox with local CLI Forums (doc collections in the Codex panel)

    Open-source alternatives such as OpenCode also exist but are not covered in this handbook.

    Platform notes

    • Apple Silicon Macs. Both agents run well as CLI tools. Claude Code and the Codex CLI are the recommended interfaces on macOS.
    • Windows. Use WSL2 (Windows Subsystem for Linux) for the best experience with both agents and the Docker-based renderer.
    • Cloud API keys. Both services require an API key or subscription. Keep your keys private and never commit them to a repository.
    • Cost. Both Claude Code and Codex require paid subscriptions. Check current pricing on each provider's website.

    What you need before starting

    1. A running PG renderer (see Setting Up the PG Renderer)
    2. An AI coding agent installed (Claude Code or OpenAI Codex)
    3. Knowledge documents loaded into the agent (see Knowledge Documents for AI Agents)
    4. A clear description of the biology problem you want to write

    Apply it today

    • Install Claude Code or OpenAI Codex following the provider's setup guide.
    • Verify your renderer is running: open a terminal and run curl http://localhost:3000/health.

    This page titled 8.1: What are AI Agents? was last modified on Fri, 27 Feb 2026 14:50:31 GMT and is shared under a CC BY 4.0 license and was authored, remixed, and/or curated by Neil R. Voss.