The Complete Procedure for Automating Contract Management with AI: Drafting, Double-Checking, Ledgers, and Automated Monitoring

Hello, I'm Himari. While teaching grooming at an animal vocational school, I also provide AI implementation support for animal-related salons.

Introduction: I had been putting off contract management for a long time

Every time a business outsourcing discussion came up, I would look at templates I found online and feel uneasy, wondering, "Is this really okay?" Managing renewal deadlines was also, honestly, vague. On the other hand, the scale of my business isn't large enough to consult a lawyer every single time.

I tried using AI to create a system for this "awkward distance from contracts," and it worked better than I imagined, so I am writing down the entire procedure.

Showing the finished product first: The 4-piece set

First, here is the big picture. What I created is a "comprehensive contract management agent" that connects the following four elements.

1. Creation: Drafting contracts through dialogue with AI
2. Verification: Double-checking where a separate AI "finds faults"
3. Ledger: Centralized deadline management using Google Drive spreadsheets
4. Monitoring: AI automatically checks deadlines and legal updates every Monday at 9:00 AM

I used Claude Code. It is a combination of custom skills, sub-agents, Google Drive integration, and scheduled execution. I will explain them in order.

STEP 1: Create a contract draft with AI

First, I set up a custom skill for legal support and created a draft for a business outsourcing agreement. There are two key points.

1. Set your "position" first
Even for the same outsourcing agreement, the clauses that should be favorable differ completely depending on whether you are the ordering party (Party A) or the receiving party (Party B). By registering your position in the initial skill settings, all subsequent drafting and reviews are conducted from that perspective.

2. Mask confidential information
I work with the other party's name and amounts left as placeholders. This is a design to minimize the information passed to the AI.

Just by filling in the blanks through dialogue, I was able to create a draft with 16 articles plus more, covering acceptance, intellectual property, confidentiality, and termination.

STEP 2: Separate the "AI that creates" from the "AI that finds faults"

This is the most important piece of know-how this time.

I have a different AI agent (a quality verifier) critically review the draft I created. The AI on the creation side tends to be lenient toward its own flaws because it is trying to "create something well." Therefore, I set up a separate AI whose only job is to "find holes."

In three reviews, there were over 15 corrections. For example—

・A one-sided clause allowing only the ordering party to terminate the contract → A fair clause allowing both parties to terminate
・Payment due date calculated from "completion of inspection" → "Within 60 days from receipt" in line with the intent of the Freelance Act
・Low cap on damages even in the event of a data breach → Confidentiality violations excluded from the cap

If I had done this alone (or as one entity?), I would have definitely missed these points.

STEP 3: Create a contract ledger in Google Drive

Contracts are not "done once created"; they are only effective if you manage their deadlines, so I create a ledger.

I created one spreadsheet in Google Drive. The columns are "Contract Name / Counterparty / Contract Type / Execution Date / Expiration Date / Automatic Renewal / 90-Day Notice Date / 30-Day Notice Date / Status / Remarks."

The trick is to include the 90-day and 30-day notice dates as columns in advance. This makes the judgment logic simple when you have it checked automatically later.

STEP 4: Set up weekly "automated monitoring"

To finish, I registered an automated execution for every Monday at 9:00 AM. There are four tasks.

① Read the Google Drive ledger
② Determine deadlines based on today (90, 30, and 7 days prior)
③ Check relevant legal amendments via web search
④ Summarize the results and notify via chat

The know-how here is to write the scheduled execution prompt to be "self-contained." Do not rely on skills or past conversations; write everything into the prompt, including procedures, file locations, notification destinations, and formats. Automated execution runs from a blank state every time, so it won't be stable unless you do this.

And in the first test run, it immediately produced results. The AI detected and notified me of a legal amendment effective January 2026, stating that the Subcontract Act has changed to the Act on Promoting Fair Transactions with Small and Medium-Sized Contractors. It was the moment the system caught an amendment I had overlooked.

What I learned from doing this: AI is not a lawyer

Finally, the most important thing.

What this system does is limited to "initial checks" and "information organization." It is designed on the premise that the final decision on important contracts must always be confirmed by a lawyer or expert. The AI review results also explicitly state, "Consult an expert for important decisions," every time.

Rather, I feel the healthiest approach is to use AI to handle the groundwork, which organizes the points of discussion when consulting an expert, allowing you to use consultation time effectively.

Conclusion

In summary, there are these four steps.

1. Set the position and masking, then create a draft with AI
2. Separate the 'creation AI' from the 'critique AI' to perform a double-check
3. Create a ledger in a spreadsheet (include a column for notification dates)
4. Set up weekly automated monitoring using a self-contained prompt

I provide AI utilization support for animal-related salons, so if you are interested in building a system like this for your own salon, please feel free to leave a comment or send me a DM.

Thank you for reading until the end.

#AIUtilization #BusinessEfficiency #Contracts #Freelance #ClaudeCode #TrimmingSalon