What Can Legal, Procurement and Compliance Teams Hand to an AI Agent?
Each build is one you can set up: what the AI agent reads, what it lists, and what a person decides. None gives legal advice.
Legal, procurement and compliance teams can hand an AI agent a first read of text: RFP risks, vendor renewal terms, changed sentences and a statement of work ledger. Each published build quotes the text, gives no verdict on it, and leaves the decision to a person. These are builds you can set up.
A first read of a long document is where hours go, and it is also a place where the document is on the table to check the AI agent against. Each build below tells the AI agent to quote the text it relies on and to say when something is not in the text. BNEDai has published a guide for each, with the instruction block to paste and a check on the first few uses. This page lists them for a legal, procurement or compliance lead choosing where a pilot could start.
A good first document task starts from text you hold
The input is text you already hold, either documents you paste or messages you forward. The output is a list or a ledger, so a reviewer reads it against the source. And the question of what to do about a finding stays with the person who is accountable for it. In every build the instructions bar the AI agent from giving a verdict on what the text means for you, and the first two also bar legal advice by name.
What the four builds share
Each uses only the text you paste or forward. Each quotes the clause, sentence or message as printed and gives its clause number or author as printed. Each writes "not found," "not in the text you gave me," or a similar line where the text is silent. And each ends by handing the decision back to you, usually as a question to ask.
Risk flags on a request for proposal
You decide, with your legal contact. The AI agent gives no legal advice and sends nothing. The RFP risk flags guide has it read the pasted text of one request and find four kinds of flag: scope wider than the fee implies, terms that move risk to you, deadlines that do not fit each other or your bid date, and questions to send before bidding. Each flag gives the clause number, the clause quoted word for word, one plain sentence on what it could cost, and a question for the buyer. It checks for payment terms, termination, limits on liability, ownership of the work, the question deadline and the start date, and writes "not found" for any that is missing.
An audit of vendor agreements before they renew
You decide whether to renew. The AI agent flags anything unclear for your legal contact, cancels nothing, and you send any question yourself. The vendor agreement audit guide has it read the agreement text you paste, with order forms and amendments, and list for each vendor the renewal date or the end of the term, the notice the agreement asks for and who it goes to, anything that renews on its own, and what you agreed to pay. Under each it quotes the clause and gives its number. It never calculates a date: where a rule sets one, it quotes the rule and leaves the date to you. It also writes one plain question you could send the vendor.
A list of what changed between two versions
You decide what each change means for you, and the instructions bar the AI agent from saying whether you are covered, compliant or exposed. In the changed sentences guide it reads only the old and new text you paste, goes through the new text from the top, and numbers every sentence that is new or worded differently, writing the old sentence on one line and the new one on the next, both quoted exactly. It paraphrases nothing and judges nothing, opens no pages and cites no law, rule or agency. It ends with questions for you to decide, such as which of your own documents use the old wording.
A ledger of requests by statement of work clause
What to do about a request is yours to decide. The statement of work ledger guide has the AI agent number every clause of a signed statement of work. You then forward each request from a Slack thread, and it files the request under the clause that matches, quoted as printed with the message and its author, or under "Not in the statement of work," and posts the ledger in the thread. It uses no other message in the channel. It gives no verdict, and does not call a request billable, in scope or out of scope.
Start with the changed sentences guide
Start with the changed sentences guide, since it reads only the two pieces of text you paste. Run it with your legal contact in the loop. The pilot playbook covers what to write down first, and what an AI agent can touch covers the permissions. The AI Readiness Index is a free 18-question check on how your team runs AI today.
Frequently asked questions
Does an AI agent give legal advice in these builds?
No. Each published build bars verdicts on the text, and the RFP and vendor builds bar legal advice by name. The AI agent quotes the text, and a person decides.
What does the AI agent do when a term is missing?
It says so. The builds write "not found" or "not in the text you gave me" instead of filling the gap.
Do these pages say what results a team got?
No. They describe how each build is set up, what it reads, what it lists and what a person decides.
Reviewed on November 20, 2026.