Agentic AI and AI agents are fun. Making all sorts of tasks or work ‘agentic’ is deeply fascinating and indeed, the results are impressive. We are making template-insertion work AI-agentic, including creating a Q&A.
The idea: AI can autonomously automate all your contracting work – with minimal, adequate human-in-the-loop (HITL).
In this short video (1:30 min), you can see what our AI agent does:
How did we make Weagree admin work agentic?
All admin work in Weagree is meticulously detailed in step-by-step tutorials. So, making Weagree admin work fully agentic was easy and straightforward:
- Claude creates scripts and skills. For every admin task, there is a tutorial with screenshots explaining step by step how any admin task is done. We asked Claude to analyse all and convert them into AI scripts and AI ‘skills’. Scripts and skills are the foundations for any AI agentic action.
- Claude follows our admin training. To enable Claude to orchestrate (identify, select and apply) the scripts and skills, we let Claude follow our admin course (Weagree admin training was designed to do all possible standard admin tasks and actions, and takes 4 hours). We gave Claude the transcripts (of the voice over of the admin training videos) and screenshots from our manuals.
- Making scripts ‘agentic’. We reviewed all admin scripts step-by-step and troubleshooted deviations with Claude, using Playwright. Playwright is for the hard work: apply all agentic actions in Weagree’s admin tooling. Playwright is what you see in the above video. https://youtu.be/_P0-tchC2d0
- Orchestrating skills (making agentic workflows). With the skills ready, we let Claude set up agentic workflows in N8N (which is the de facto AI platform for agentic workflows). N8N will orchestrate invoking the scripts and skills as appropriate. As workflow platform, N8N is the place to trigger human-in-the-loop (HITL) to review and approve intermediate results and coordinate next agentic steps.
- Gen AI to agenticise orchestration. Claude (or any other LLM) will review the uploaded model contract to identify which skills are needed (when and where), and N8N will process them all. Claude will identify the document styling, identify who is who (e.g., distributor, service provider, customer, licensor), identify cross-references, definitions and schedules, the Q&A-points, formulate the Q&A questions, formulate the Q&A-dropdown variables, link everything to CLM metadata.
The AI scripts and AI ‘skills’ for questionnaires are also ready to go (we only need to hang them in OpenClaw’s N8N agentic workflows). That’s for next time.
What steps does an AI admin agent take?
What does an AI agent, when model contracts are agentically automated:
- Remove cover page, table of contents, etc. (they will be inserted via template settings later)
- Analyse the contract styling and clause indentations
- Upload the model contract (using existing Weagree admin tooling)
- Apply level-styles and formatting (using existing Weagree admin tooling)
- Identify where definitions are listed
- Remove standard clauses (to insert clause-library clauses later)
- Insert clause-library clauses (replacing
- Tag definitions (so that none is unused and none non-defined)
- Tag schedules and annexes (so that all appear if referenced, and all referenced correctly)
- Automate all cross-references
Automating your entire knowledge base of model contracts has never been as simple. Weagree AI agents do all your admin work: with AI agents. Watch the video to see it yourself. What’s next?
- In video 2 (forthcoming), we will demonstrate how a questionnaire is inserted by Weagree AI agents.
- Video 3 (also forthcoming) will show how AI agents can modify automated contracts (template maintenance work).
Why contract automation in Weagree?
When drafting a contract, you don’t want AI to identify and process a Q&A using thousands of tokens and datacentre capacity, and even risk hallucination. Things that are so predictable that they better be ‘merely’ automated. You do not want an AI to answer contract-creation questions that you need to answer yourself.
Did you think that AI-drafted contracts require less questions? Think again: your AI will either invent answers (hallucination) or ask the same questions as you would automate in Weagree.
No matter how impressive AI can draft contracts (even based on your own templates), AI is a heavily exaggerated tool for drafting what is better left ‘automated’. (Buying a bread driving your Ferrari is ludicrous, even if you don’t care about the oil prices.) On your own model contracts, you want an AI agents to maintain your templates and be sure all (meta)data are collected and properly reflected.
