Skip to content

What Are Coding Agents?

From Chatbot to Agent

In the AI Essentials section, we covered how to use AI chatbots well: picking the right model, writing strong prompts, organizing projects. That is genuinely valuable work. But chatbots have a fundamental limitation: they can only talk.

A chatbot can draft a contract summary — but you have to copy-paste the contract in, copy the summary out, and figure out where to put it. A chatbot can suggest how to organize your files — but it cannot actually move them. It can write code — but it cannot run it.

A coding agent is different. It is an AI that can take action on your computer. It can:

  • Read your files — open a 50-page PDF, a folder of case documents, or an entire directory of course materials
  • Run commands — execute searches, create spreadsheets, process data, send emails
  • Build things — generate documents, create websites, set up databases, write and run programs
  • Remember context — maintain a persistent understanding of your projects, preferences, and working style

Think of the difference this way: a chatbot is like calling a very smart colleague on the phone. A coding agent is like having that colleague sit at your desk with access to your files, your email, and your calendar — and the ability to act on what they find.

Models, apps, and harnesses

A useful framework from Ethan Mollick: the AI you experience is actually three things stacked together.

Layer What it is Example
Model The underlying intelligence — the AI brain itself Claude Opus, ChatGPT's reasoning models, Gemini Pro
App The interface you interact with Claude.ai website, ChatGPT mobile app
Harness The system that lets the model take actions — use tools, read files, run code Claude Code, OpenAI Codex, Co-work

This distinction matters because the same model behaves very differently depending on its harness. Claude Opus in a chat window can only respond to what you paste in. Claude Opus in Claude Code can read your files, search your email, write and run scripts, and execute multi-step workflows autonomously. Same intelligence, dramatically more capability.

When people say "I tried AI and it was not that useful," they are almost always describing a model in a basic app with no harness. The shift from chatbot to coding agent is primarily a shift in harness — and that is what unlocks the capabilities described on the rest of this page.

Why Now?

Coding agents are not new in concept, but until recently they required significant programming experience to use. What changed in 2025-2026 was accessibility:

  • Natural language control. We can tell agents what to do in plain English. No programming syntax required.
  • Reliable reasoning. The underlying AI models became good enough to handle multi-step tasks without getting confused or lost halfway through.
  • Safety guardrails. Modern agents ask for permission before taking destructive actions, so we can experiment without fear of breaking things.
  • Documentation and community. Tools like Claude Code invested heavily in guides, tutorials, and support for non-technical users.

The practical result: a law professor who has never written a line of code can install a coding agent and have it do useful work the same afternoon. That was not true two years ago. It is true now.

What We Cover in This Section

  • The Agent Landscape

    A survey of the major coding agent tools available today — what each one does, how they compare, and why skills transfer across all of them.

    See the landscape

  • Why Claude Code?

    Why this guide uses Claude Code specifically, what makes it well-suited for legal work, and an honest account of its downsides.

    Our choice explained

  • What Agents Can Do

    Concrete, law-specific examples of agent capabilities — from contract analysis to course design — described in plain language.

    See the possibilities

  • Limitations & Risks

    What agents get wrong, where they pose professional risks, and the responsible use framework every lawyer needs to understand.

    Know the boundaries

The Key Insight

We do not need to become programmers to use these tools. We need to become good at describing what we want — which, as lawyers and legal scholars, we already do for a living. The skills that make someone good at drafting jury instructions, writing case briefs, or structuring a law review article are the same skills that make someone effective with a coding agent.

The rest of this section explains the landscape, makes the case for our chosen tool, shows what is possible, and — critically — explains what can go wrong.