What does it entail: AI is taking over legal work, is it making legal professionals redundant? Let me tell you all the secrets of building your own Legora or Harvey.
Weagree becomes AI agentic
We are making Weagree almost entirely ‘AI agentic’. If a customer wants to:
- Automate their templates in Weagree – an AI agent should do that.
- Configure Weagree for certain types of users or use cases – an AI agent should do that.
- Search a contract by asking in normal human language and find it in Weagree CLM – an AI should do that.
- Convert their model contracts into contract-review playbooks (with fallback clauses), an AI agent should do that.
In other words, a fully AI agentic Weagree should deploy AI to support customers, solve our customers’ legal-tech challenges, advise them about how-to-drive-adoption, develop legal-AI solutions and do entire sales-cycles.
Let’s build it:
Can everyone build an agentic AI platform?
Over 2024/2025, everyone has seen how Gen AI can boost an organisation, but how does agentic AI extrapolates that? What is so special about Legora or Harvey that law firms are willing to spend such tremendous subscription fees?
Developments are going too fast – even for Legora or Harvey – to develop a sustainable platform of their own. Even if they did, leading open-source platforms will be introducing new AI-capabilities and also Legora and Harvey cannot permit being on an entirely self-developed ‘tech stack’. We saw it last week, when Claude introduced a Word add-in that is better than Spellbook’s Word add-in (or ours).
Legora and Harvey must have taken the de facto market-leading Agentic AI tools, package them together and dressed it all up with:
- A glossy interface,
- Some extra configuration-options (translating generic settings into legal-practice jargon), as well as
- ‘Skills’ and ‘agentic workflows’ specific for legal tasks as they predefine themselves, and
- User management and API integrations.
Let’s 'vibe-code' agentic AI
Most or all of agentic AI consists of open-source connected tools. So, what are the leading agentic AI platforms that would enable Weagree to automate templates by AI, configure Weagree autonomously, or convert model contracts into playbooks with appropriate controls for lawyers in the loop?
Most or all of agentic AI consists of open-source connected tools. So, what are the leading agentic AI platforms that would enable Weagree to automate templates by AI, configure Weagree autonomously, or convert model contracts into playbooks with appropriate controls for lawyers in the loop?
Here is the recipe (that you can deploy yourself as well):
- OpenClaw is your core for agentic AI. It connects Whatsapp, Slack, Signal and your e-mail (soon, it will include Teams and Outlook). It links up to all files that you open up (and when you grant OpenClaw access to a file or folder, it will probably ‘index’ it so as to respond fast to any incoming request. OpenClaw is the core that also connects the AI models being called by agents.
- N8N for easy workflow configuration. Although OpenClaw enables the creation of workflows, in N8N building workflows is more intuitive (N8N provides a canvas on which all sorts of ‘nodes’ can be created, configured and connected). More importantly, N8N provides a marketplace for thousands of preconfigured template-workflows.
- Paperclip is the orchestrator of AI agents. Where OpenClaw can be seen as an AI agent (or a small team of AI agents) with some dedicated task or role, Paperclip is the AI organisation of all agents. Paperclip structures the AI agents in an org chart, and gives control over the AI costs.
- Playwright is the in-browser AI agent: everything that must be performed in a browser, like template insertion work or configuring Weagree CLM (or conducting sales on LinkedIn), would be done by Playwright: a sort of invisible layer in your browser (behind the user interface and the internet) that ‘sees’ everything on screen, and can even perform the actions that a human would do. Also this is something OpenClaw could do, but Playwright is more specialised and has a deeper ‘understanding’ of what it does in your browser. More importantly, Playwright is better equipped to create user-scripted (AI) ‘skills’ related to browser-actions and enables human-in-the-loop (appropriate control over AI-agentic actions).
- Postgres, a vectorised database that memorises how users prefer to work and ‘remembers’ what was done previously and how. Postgres is where agent-configurations, skills and workflows, and OpenClaw’s capabilities are stored.
- Tailscale for IT security (controlling access rights).
- AI models for the gen AI work: helping you step-by-step in installing, configuring, connecting, trouble-shooting and testing the above; and after launch, performing their in-app tasks and role, and answering your chat-questions (not everything requires advanced-thinking of gen AI – many tasks can be done by models like Ollama). You may deploy several AI models: Ollama can do straightforward tasks, is free and can be installed on your computer; and some AI models are better in searching and others in writing.
These leading agentic AI technologies are open source (entirely free) and available for you and me. Setting up your own solution takes a week or so. If you use Claude as your worker, she will do all the hard work for you.
Cooking it yourself
Many independent legal professionals set up their own agentic AI structure. GitHub is the (free) place where complete preconfigured software solutions can be found. Connect your GitHub account to Playground (a software development application) and follow the above recipe for agentic AI to set up your own legal-tech stack.
For Weagree, the agentic AI-tech stack consists of OpenClaw, N8N, Paperclip, Tailscale and Postgres, with Playwright alongside. For a legal department of up to 10-20 lawyers, OpenClaw, Postgres and Tailscale would suffice (N8N might be an appealing add-on for its user-friendly workflow configuration).
Claude (or any other AI) will assist in connecting it, configuring all, and fill it with content. As each of the above components will be connected to AI, the in-app AI chat will help you do whatever you want to achieve in that component; in N8N, the creation and finetuning of agentic legal workflows, in Paperclip the creation of ‘personas’ and in each of OpenClaw, N8N, Paperclip and Playwright the ‘skills’ that your AI will use in drafting a contract clause, inserting a template, creating a playbook or configuring Weagree.
Yes you can!
…create your own Legora or Harvey! The legal tech world is developing so fast, that with the right agentic-AI legal-tech stack, everyone can create an equivalent of Legora and Harvey. And that is actually what happens. Legal-tech influencers publish their ‘skills’ and templated workflows (for free). It takes a few clicks to deploy them… in your own tech stack.
Any idea how to implement AI and move towards agentic AI?
