What if lawyers could build their own legal tech tool in a weekend – without a software team or a computer science degree? That’s the promise of “vibe coding,” a term coined by AI researcher Andrej Karpathy to describe a new way of programming: instead of writing code line by line, legal user can describe what he want in natural language, and let an AI generate the code. In Karpathy’s words, “there’s a new kind of coding… where you fully give in to the vibes… and forget that the code even exists.” This might sound like science fiction or hype, but lawyers from big-firm to in-house counsels are already doing it – prototype-building legal apps by simply chatting with AI assistants. In this article, we’ll demystify vibe coding, show real examples of lawyers who’ve built useful tools with it.
DISCLAIMER: This article does not suggest that the legal profession is being replaced, nor that traditional coding is obsolete. Rather, it highlights emerging options in LegalTech that expand who can participate in building legal tools. Lawyers, developers, and multidisciplinary teams each continue to play essential roles as the market evolves.
I. What Is “Vibe Coding,” Anyway?
Vibe coding is programming by using AI and “feel” rather than formal code syntax. Think of telling an assistant, “build me a simple app that does X,” and it generates the working code. You iterate by saying “hmm, make that button bigger” or “extract the dates from this document” and accepting the AI’s changes. Karpathy introduced the term in early 2025 to capture this almost conversational style of development.
Importantly, vibe coding is not just any AI coding. It doesn’t mean every time you use GitHub Copilot or Stack Overflow you’re vibe coding. Karpathy’s point (and Simon Willison’s, in a clarifying blog post) is that vibe coding implies speed and nonchalance – building quick-and-dirty prototypes, not meticulously engineering every part of the system. In Willison’s words: Using LLMs responsibly to generate code that you then carefully review is just modern software development, not vibe coding. Vibe coding is more like an experimental jam session with an AI pair-programmer – fun, creative, and definitely not for high-stakes missions.
So why are we talking about it in a legal context (where stakes are high)? Because it turns out for many low-stakes legal tasks and prototypes, vibe coding “just works.” Lawyers who can’t code a lick have used it to whip up solutions to real problems in hours. Before we dive into those examples, let’s set the stage for why this is happening now.
II. The Case for Exploring Vibe Coding in Today’s Legal Landscape
- AI Capability Leap: Recent large language models (LLMs) that are surprisingly good at generating coherent text – including code. Around late 2022, models like GPT started reliably producing working code from plain English prompts. By 2023, GitHub Copilot was completing lines and even entire functions for developers. By 2024, tools like OpenAI’s Code Interpreter (now ChatGPT’s Advanced Data Analysis) could generate and execute code to solve problems. In short, the tech reached a tipping point. Karpathy noted that “LLMs… are getting too good” – good enough to trust for a weekend project.
- No-Code/Low-Code Maturity: Parallel to AI, the no-code movement has been blossoming. Legal teams have been using no-code automation for years (think contract assembly in Word, workflow in tools like SharePoint or Onit, expert systems in Bryter, etc.). People became comfortable dragging and dropping logic without “coding.” Vibe coding is a natural next step: instead of dragging blocks, users are typing requests to an AI. It’s different technology, but the mindset (“I can build this myself without a developer”) was already taking root. As one legal tech blogger put it, until recently lawyer-built software meant no-code or low-code platforms, but now LLMs are introducing “a new layer of building altogether” as pointed out by Alex Herrity in his article The Tools Lawyers (Might) Build For Themselves.
- Precedent of Success: We now have success stories that inspire others (I’ll share a few in the next section). When a Clifford Chance associate or a solo GC posts on LinkedIn that they built a contract review bot over the weekend, it lowers the psychological barrier for everyone else. There’s a growing community (on LegalTech LinkedIn, in innovation groups, etc.) of “lawyer-makers” sharing what they built. In other words, the fear of being first is gone. If 79% of lawyers were already experimenting with Gen AI in some form by late 2024 as pointed out in article What Are the Data Privacy Implications of Using AI Tools with Confidential Client Information?
To sum up: the tech is ready, the need is there, and the cultural shift has begun. Lawyers are increasingly tech-savvy and willing to try new things post-COVID. Vibe coding hits at an opportune time when the profession is looking for efficiency and innovation.
III. The New “Legal Tech Stack”: What Can Lawyers Use to Vibe Code?
One might be thinking, “This sounds cool, but how would I even do it?” Lawyers don’t have to build an AI coding setup from scratch – there’s an emerging toolbox of products that enable vibe coding. Here are the main categories and examples in each:
- AI Coding Assistants (IDE Plugins): Having even basic coding skills, with tools like GitHub Copilot (by OpenAI/GitHub) or Amazon CodeWhisperer one can turbocharge development. Copilot plugs into VS Code, JetBrains, etc., and autocompletes code and writes functions based on comments. User describe what he wants in a comment (e.g. “// function to parse dates from contract text”) and Copilot suggests code. It’s trained on billions of lines of code, so it often knows what to do. This is more for traditional coding workflows – and indeed many law firm tech teams are using Copilot to accelerate their internal dev – but it’s one piece of the puzzle. It can turn a junior developer (or “citizen developer”) into a much more capable one. Copilot now even has a voice-based “Copilot X” that can talk you through code or suggest tests. Think of these as AI sidekicks for building legal apps. They reduce the heavy lifting of syntax and searching for snippets.
- Conversational “Build Code by Chat” Tools: A step further are tools like Cursor or Replit Ghostwriter that provide a full development environment centered on conversation. Cursor, for instance, is an AI-first code editor where user can open a sidebar and literally chat with its codebase. User might giva a task, “Hey, find where I define the term ‘Client’ in this repo,” and it will find it. Or “Write a function to anonymize personal data in this document text,” and it will generate it. Cursor’s own model (“Composer” with model Sonnet) was cited by Karpathy as enabling his vibe coding spree. It’s basically an IDE that wants you to vibe code. Similarly, Replit’s Ghostwriter lets user describe an app you want and it spins up the files and code in your browser. Even Visual Studio Code has extensions where one can highlight code and ask in plain English for modifications. These tools blur the line between coding and chatting. For a lawyer, it means he can start expressing what he needs (“I want a form that does X, then Y”) and the tool scaffolds a project for you. Early users say it feels like collaborating with an eager junior developer – sometimes great, sometimes mistakes, but a huge speed boost.
- No-Code/Low-Code Platforms with AI: Even without any coding background, lawyers can still build effective solutions — many established no-code platforms now offer integrated AI capabilities to support non-technical users. Retool, a popular internal app builder (used for making admin dashboards, etc.), has added AI features where you can type what you want and it configures the app. Airtable (which many legal teams use as a pseudo-database) introduced Airtable AI to generate formulas and summaries. Zapier now has natural language automation (“tell Zapier in English what workflow you want to automate”). And then there are niche legal no-code tools: e.g. Checkbox AI allows building guided workflows and recently launched an AI assistant to create those workflows from a prompt. Similarly, Josef (legal bot builder) added GPT for drafting content. So users can vibe code within these platforms: describe their intake process, and let the system draft the workflow, which you can then tweak via their visual interface.
- Generative Prototype Tools: These are worth separate mention because they’re the darlings of the “vibe coding” movement outside legal. Vercel’s “v0” (pronounced “v-zero”) is one such tool. Vercel (the company behind Next.js) built v0 to let anyone create web app frontends by typing requests. For example, “Generate a simple webpage for a contract review checklist with a form for contract text on the left and an AI summary on the right.” V0 will produce the code for that layout and even hook in an LLM API if you specify. It’s like telling a junior dev, “make it pretty and functional,” and they come back with HTML/CSS/JS ready to deploy. Another one, Bolt, can handle full-stack generation – it will create the frontend and backend logic and database integration just from your prompt. These are more experimental, but people have used them to prototype things like mini-CRM systems or internal tools extremely quickly. Imagine: a legal ops manager could describe a basic contract management tool and get a working prototype to show the team – without waiting months for IT. These tools often require technical cleanup, but they drastically shorten the journey from idea to something clickable.
- Domain-Specific Legal AI: Lastly, I’ll note the rise of “legal Copilots” – like Harvey, Casetext’s CoCounsel, etc. These are more about using AI within legal workflows (e.g. ask Harvey to review a contract and highlight issues). They’re not vibe coding tools per se, but some allow customization. For example, CoCounsel has skills that can be configured via natural language, and Harvey is working on letting firms “train” it on their own templates and preferred clauses. It’s not the same as building a new app, but it is a form of building automation by guidance. If user tell Harvey “compare this contract to our playbook and mark deviations,” you’ve essentially created a custom review workflow for your team. Over time, these domain AIs might expose more “builder” functionality – so a lawyer could string together a series of AI analyses into a mini automated workflow. We’re heading there.
Bottom line: There are onlt few options mentioned above. Lawyers now have an array of tools to turn their ideas into reality. Whether users are comfortable writing a bit of code or you want pure no-code, vibe coding can happen at whatever layer you operate on. The key is the mindset of describing what user want and iterating quickly, rather than waiting for a formal software project. Next, let’s see how some of your peers have put these tools to use in the real world.
IV. The New “Legal Tech Stack”: What Can Lawyers Use to Vibe Code?
To make this concrete, here are snapshots of how legal professionals – not trained software developers – have successfully vibe-coded or no-coded solutions in the past year:
“SpellPage” Contract Editor – Built by a BigLaw Associate
Jamie Tso, a senior associate at Clifford Chance, received widespread attention among legal professionals on LinkedIn in 2025 by sharing the AI tools he built in his spare time. He’s an M&A lawyer, not a programmer – but when faced with repetitive tasks (like comparing prospectus language to regulatory checklists), he started experimenting. Using the firm’s Microsoft Copilot access and some Python, Jamie created a contract analysis assistant that could redline a document in real-time based on his spoken or written instructions. Essentially, he had a Word-like editor where he could say, “Highlight any indemnity clauses that deviate from our standard,” and the AI would do it. He named parts of this system “SpellPage” (inspired by a novel-writing app UI, funnily enough).
Anson’s GPT-Powered Word Add-In – Built by an In-House Counsel
Anson Lai, A commercial counsel who handles technology contract negotiations by day and develops AI-powered contract review tools in his own time. Frustrated with constantly copying and pasting contract text into ChatGPT for analysis, Anson decided to bring the AI to his documents. In late 2025, he vibe-coded a Microsoft Word add-in that basically embeds an AI assistant in the Word ribbon. Highlight a paragraph and ask, “Does this meet our standards?” – it will answer in a sidebar. Or, “Redline this clause to be GDPR compliant” – it can suggest edits. It even has a chat mode where you can ask, “What are the key risks in this contract?” This tool rivaled some startup products on the market – and he built it in a few weeks, on nights and weekends. How? He used Google’s AI IDE (Project Antigravity) and the latest GPT-4 model (Gemini) via API, essentially talking to the IDE to generate the plugin code.
“Artifex” AI Assistant – Built inside a Law Firm (30% staff usage)
Let’s one thinks vibe coding is just for solo tinkerers, here’s a story of a firm institutionalizing it. Buchanan Ingersoll & Rooney, a large firm, didn’t wait for vendors to hand them solutions – they built their own AI platform called Artifex. It started in 2022 when the CIO and IT Director set up a sandbox to play with generative AI (initially just OpenAI’s API). They were cautious: as security experts, they ensured it was a closed environment and that no client data would leak. Over 2023, they kept iterating Artifex in-house. It grew from a nifty demo to a sophisticated internal web app that lawyers can log into for various AI-powered tasks. What tasks? For example, summarizing documents – attorneys can drop a 100-page regulatory guidance PDF in, and Artifex will produce a synopsis, saving hours. Or comparing versions of a contract – something associates often do manually – Artifex can line up two drafts and highlight the differences and even flag potentially concerning changes. It also has a Q&A feature: lawyers upload a document or case and ask questions to get targeted answers (like “what’s the indemnity cap here?”). Adoption has been strong: within a year, about 30% of the firm’s 450 personnel were actively using Artifex. This project shows a path for others: start small in a sandbox, prove the concept, then invest more once you see the value – all while keeping IT and risk folks in the loop.
Compliance Playbooks Turned Apps – No-Code at a Global Firm
Another flavor of lawyer-led building is using no-code platforms to encode legal expertise. A notable example comes from Linklaters in Europe. In 2022, Linklaters formed a partnership with a no-code automation platform (Bryter) to enable its lawyers to create client-facing tools. One early success was a “Dawn Raid App.” For the uninitiated: a “dawn raid” is when a regulator (like an antitrust authority) shows up without warning to search a company’s premises. There’s a specific checklist of things the company’s legal team should do immediately. Linklaters’ antitrust lawyers had a detailed playbook for this, covering many countries. Instead of a PDF memo, they built an interactive app: the user selects the jurisdiction and answers a few questions, and the app provides step-by-step guidance tailored to that scenario. It even tells the user when to call which Linklaters lawyer, etc.
These case studies show a range of approaches – from pure LLM vibe coding to structured no-code apps – but the common theme is lawyers identifying a problem and rapidly building a tool to solve it. None of these required writing thousands of lines of Java or a million-dollar budget. They required curiosity, some free time, and perhaps a supportive boss or two.
V. The Risk Checklist: Keep It Legal (and Safe) When Vibe Coding
It’s ironic: As legal professionals, we are trained to assess risk with precision — and while this article explores new approaches to innovation, it does not overlook the responsibilities that come with them. Below is a summary of key risks associated with vibe coding in legal contexts, along with practical strategies for mitigating them effectively.
- Confidential Data Exposure: This is number one. If user use cloud AI services (and he likely will, unless he is running an LLM locally on his laptop), whatever he send might leave your secure environment. We saw some high-profile scares in 2023, like lawyers pasting client memos into ChatGPT and – oops – that data living on OpenAI’s servers. In fact, ABA Formal Opinion 512 (July 2023) directly tackled this: it requires lawyers to “understand if AI tools are self-learning (i.e., use your inputs to train) and to get client consent before sharing confidential info”.
- Hallucinations & Inaccuracies: LLMs can “hallucinate” – make stuff up – in very convincing ways. If your vibe-coded app provides advice or analysis, there’s a risk it outputs something that’s flat wrong or even fictitious (like citing a case that doesn’t exist). Mitigation: keep a human in the loop. Think of the AI’s output as a draft. Also, whenever possible, have the AI show its sources: e.g., if it’s answering a legal question, program it to provide the clause or reference it based the answer on (some LLMs can do this via retrieval augmented generation).
- Unauthorized Practice / Giving Legal Advice via Software: If user build a client-facing tool, remember that he still needs to be on the hook for its output. Let’s say lawyer offer a public web app that assesses whether an employee classification is likely correct – that could be considered legal advice. To avoid any issue, it is necessary to always include disclaimers and limitations and be transparent. E.g., “This tool is for informational purposes, not a substitute for legal advice. Consult counsel for advice on specific situations.” Also, avoid too definitive of language in client-facing tools.
- Security & Access Control: Think like the IT department for a moment. If one spin up a quick web app to upload contracts and analyze them, where is that hosted? Could someone else access those uploaded docs? A common mistake in DIY apps is neglecting security (e.g., leaving an S3 bucket of uploaded files open). Even if users are just using a tool internally, it is necessary to ensure proper access controls. If it’s a local script, fine. If it’s a web app, even internal, have a login or limit it to our VPN. We also need to consider penetration testing if it becomes widely used – maybe ask IT to run basic security scans on it.
- Maintenance & Reliability: Today’s cool prototype can become tomorrow’s critical process – and if it breaks, who fixes it? Lawyers move roles, summer interns leave, etc. If you build something that people start relying on, make sure there’s at least a plan for maintenance. Document the basic “how” of the tool. Even a one-page readme is better than nothing.
- Bias and Fairness: If our tool makes decisions (like triaging cases or recommending contract clauses), we are required to be mindful of bias. AI can reflect biases in training data. For example, an AI trained on past litigation outcomes might undervalue cases involving certain jurisdictions or plaintiff types if the data was biased. In contract drafting, an AI might favor vendor-friendly language if it saw more of that. So, sanity-check for any skew. Since as lawyers we operate under fairness and anti-discrimination duties (and increasingly regulations like EU AI Act or FTC guidance), ensure your vibe-coded solution isn’t inadvertently causing a bias issue.
For decades, we’ve consumed technology built for us (often without our input, leading to mismatched solutions). With vibe coding and no-code tools, more lawyers can become co-creators of their tech solutions. That’s a powerful shift. It means the legal tech landscape might see more “bottom-up” innovation – the new workflow at your firm next year might come not from IT, but from a paralegal or junior associate who built a tool and shared it.
