Most teams that handle contracts run into the same routine. A new NDA from a client or vendor needs to be reviewed by a legal team. They read through it, walking down the same list of risks they always check: Term length, Governing law, Liability cap, Non-solicit, Termination. They mark what's acceptable, flag what isn't, write up the findings and send them back. It takes a lot of time and it happens several times a month.
From the inside most of the process is mechanical: going down a checklist that already exists and matching each item against the document. We wanted to see if we could solve it.
We created an AI agent that extracts the text from the uploaded NDA, loads the checklist your team has defined, then goes through the checklist one risk at a time. For each risk it searches the document for the relevant clause, evaluates whether the language meets the criteria and records the finding with a citation back to the original text.
The lawyer's job doesn't disappear. The agent only checks risks that are on the checklist. The lawyer reviews the agent's findings, verifies borderline calls, and flags anything non-typical or outside the checklist that still needs attention. The agent handles the routine; the lawyer handles the exceptions.
Every company has its own NDA policies. A 5-year term might be fine for one team and a deal-breaker for another. A liability cap that's standard in one industry is unworkable in another.
Generic AI doesn't know your rules. Ask it to "review this NDA and tell me what's risky" and the output will be inconsistent. Some real issues get missed. Some non-issues get flagged.
We started with the lawyers, not the model. Together with the legal team, we wrote down the actual rules: every risk they check for, the threshold for each, and the criteria that make a clause acceptable.
The LLM then does the narrow part: it takes the checklist, finds each risk in the document, and marks it compliant or non-compliant with the clause quoted as evidence. The agent doesn't decide what counts as a risk. It checks the document against the rules the legal team already wrote down.
The agent is orchestrated in n8n, with Slack as the entry point, Google Drive as storage for the checklist and OpenAI's GPT as the model that does the analysis. For the legal team there's no new tool to learn: they upload the NDA to a Slack channel and the agent's reply lands in the same thread.
The stack is flexible by design. We can adapt the architecture to whatever the client already runs: another orchestrator instead of n8n, Teams instead of Slack. The value lives in the workflow, not the specific tools.
The pattern isn't specific to NDAs. The same approach works for MSAs, DPAs, vendor agreements, and any document review where a checklist already exists in someone's head. If you look at your own processes, there's probably several similar tasks.
This agent is running in production at T1A, reviewing our own NDAs every day.
If this looks like tasks you're dealing with, get in touch — we can build you an agent like this in a few weeks.
If you're curious about other AI agents we're building at T1A: Contract Intelligence Agent: Making Contract Knowledge a Shared Resource — the inverse problem, applied to contracts you've already signed.
