AI Agents for Lawyers

Glossary

Vibe Coding — Definition for Lawyers

Vibe coding means building software by describing what you want to an AI tool in plain language and iterating on the result. What lawyers build with it, and its limits.

By Dr. Niklas SchmidtUpdated 15 November 20263 min read
On this page
  1. How it works
  2. What lawyers build, and what they should not
  3. A specification that gets better results
  4. Limits and data-handling cautions
  5. Where this leaves you

Vibe coding is building software by describing what you want to an AI model in plain language, running what it produces and asking for changes until it works, without writing or necessarily reading the code yourself. For lawyers it turns a spreadsheet-sized irritation into a working tool in an afternoon. It does not turn a lawyer into a software engineer, and the difference matters as soon as client data is involved.

How it works

The loop is short. You describe the tool, the AI generates the code and usually a live preview, you try it, and you describe what is wrong. Repeat. The tools range from chat assistants that produce a self-contained web page, through browser-based app builders that also host the result, to coding agents such as Claude Code that work on files on your own machine.

What makes it work for lawyers is that the hard part of a small legal tool is rarely the code. It is knowing what the rules are, which edge cases matter and what the output must look like. Lawyers have that knowledge; the AI supplies the syntax.

What lawyers build, and what they should not

Good fitPoor fit
A limitation-period or fee calculator for one jurisdictionA firm-wide docketing system
A client intake questionnaire that routes to the right teamAnything that gives legal advice to clients without review
A decision tree for a regulatory threshold testA system of record for client money or conflicts
A tool that sorts signature pages or renames files in bulkSoftware others depend on that only one person can maintain
A clause comparison or side-letter tabulator for one dealIntegrations with the DMS or billing system built without IT

The pattern on the left is a bounded tool that does one thing, used by people who can tell when its output is wrong. The decision-tree playbook and the client-intake playbook each walk through one build from blank screen to working app.

A specification that gets better results

The most common failure is prompting with a one-line idea. Write the spec as a user experience first:

Build a single-page web app for internal use at a law firm.

User: an associate checking whether a share transfer needs notification.
Inputs: target's country, acquirer's country, percentage acquired,
sector (dropdown: defence, energy, telecoms, media, other).
Logic: [paste your decision rules as numbered if/then statements]
Output: "Notification likely required", "Not required on these facts" or
"Refer to partner", with the rule that triggered the result quoted.
Constraints: no data leaves the browser; no external calls; no login.
Include five test cases with expected outputs and a button to run them.

The test cases are the part lawyers skip and should not. They are how you know the tool applies your rules rather than plausible-looking ones.

Limits and data-handling cautions

Three risks recur. Data handling: an app builder may host your app and store what users type, so prototype on synthetic data and check where anything is stored before real information goes near it. Correctness: the code can look right and implement the rule wrongly, so verify against cases you know. Maintenance: a tool nobody else understands becomes a liability when its author moves on. The build-or-buy guide sets out when a vibe-coded tool should be replaced by a supported product, and the agent security guide covers where coding agents stop being safe to run on firm machines.

Where this leaves you

Pick one repeated task that annoys you, write the specification and test cases on paper, and build it with synthetic data. If it survives a week of real use by colleagues, decide with IT whether it stays. Module 5 of the AI Lab for Lawyers course covers vibe coding calculators, questionnaires, decision trees and client-intake apps without code.

Frequently asked questions

Do I need to know how to program to vibe code?

No, and that is the point of the term. You describe the tool, the AI writes the code, and you judge the result by using it. What you do need is the ability to specify precisely what the tool should do, to test it against cases where you know the right answer, and to recognise when a problem is beyond what you can safely maintain.

Can I use a vibe-coded app with client data?

Not without the same scrutiny you would give any software vendor. A prototype built in an afternoon may store data in places you did not choose, log inputs, or rely on a hosting service whose terms you have not read. Build and test with synthetic data, then have whoever is responsible for IT and data protection review the app before real client information goes near it.

Put this into practice

AI Lab for Lawyers: a live, hands-on course for practising lawyers

4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion