What you'll learn:
Why NDAs beat every other contract as a first automation
How to turn the checklist in your head into rules an AI can test
The five pieces you need to set this up, whatever tool you use
A salesperson types one line into #nda-tracker. The counterparty gets a link, fills in their own details and signs. I find out when the signed copy lands in the channel, already logged.
That's what most NDAs should look like. Nobody writes an email.
NDAs are the best contracting function to automate first. Here's the case.
Why NDAs first
Three reasons.
Low risk. Most NDAs are routine. In tech, the exceptions are narrow: corporate transactions (M&A and fundraising), sharing personal data and co-developing a product together. Outside those, the downside is small.
Few variables. An NDA has a limited set of provisions. Very little changes from one to the next.
You already have a playbook. Maybe it's in your head. Maybe it's on a sticky note. Either way, you know what you accept and what you change. You're encoding a decision you've already made, not inventing one.
Most other contracts have more provisions and more judgment per page. NDAs give you the most repetition for the least judgment. That's what a first workflow should look like.
Keep NDAs out of sales and legal
We run NDAs in their own channel, #nda-tracker. Not in sales. Not in legal.
Not every NDA becomes a customer. Plenty go nowhere. In the sales channel they clutter the pipeline. In the legal channel they bury everything else. A separate channel keeps the volume out of the way and gives us one place to look.
It's also the front door. If an NDA shows up in a DM, an email or a deal thread, the first move is pointing the person to #nda-tracker and getting the document posted there. One door, one process.
Two kinds of NDAs
There are two kinds, and they're different jobs.
Their paper. The counterparty sends their NDA. We review it against our playbook, redline it and log it. This is where the judgment lives.
Our paper. We send ours. There's nothing to negotiate, so there's no review. The salesperson requests it in the channel and a self-serve form does the rest. Legal only checks the signed copy for completeness afterward.
Both come through the same door. Parts 2 and 3 cover the tools that run each one.
The playbook is a set of tests
Ours has 17 rules. Each rule has specific criteria, and a rule passes only if every criterion is true. One miss fails the whole rule. Every rule also has a severity. Eleven are major and six are minor, which tells me what to look at first and what I'll trade in a negotiation.
"Review the survival period" is a prompt. A test looks like this:
Rule: Confidentiality survival (major)
General information survives no longer than a fixed number of years.
No perpetual obligations for general information.
Trade secrets stay protected for as long as they remain trade secrets.
The AI doesn't guess what I care about. It checks each statement and answers yes or no.
I'm not sharing our real positions, and the example above is illustrative. Yours should come from your own risk tolerance anyway. Copying mine would mean borrowing someone else's judgment. If a rule doesn't change an outcome, it doesn't belong in the playbook.
The register
Every NDA gets logged in one register with fixed fields. One rule keeps it useful: if something isn't known yet, we write "to be confirmed" instead of leaving it blank. A blank looks like nothing happened. A "to be confirmed" is an open item someone can chase.
Six months from now, when someone asks whether we have an NDA with a company, the answer is one lookup. That's what turns a fast reviewer into a system.
Where the human stays in
Corporate transactions, personal data and co-development aren't low-risk NDAs. They shouldn't get an adversarial redline and a quick sign-off.
So one rule in the playbook exists for exactly that. When a third-party NDA is for one of those purposes, the rule fails, the automation stops and the NDA goes to legal for individual review. Nothing gets redlined and sent back from the workflow.
The AI can spot the pattern. It can't tell when a conversation is quietly turning into a partnership. That call stays with a person.
For our own NDAs, we trust the business to use the standard form for standard conversations.
What you need
Five things, whatever tool you use:
A dedicated channel for NDA intake, with one rule: everything comes here
A playbook written as pass/fail rules with severity, plus your fallback positions and one rule that stops and escalates the NDAs you won't automate
An AI tool that can hold the playbook, read a document and draft a redline
A standard NDA of your own, sent through a self-serve e-sign link
A register with fixed fields, updated after every event
The tool is the easy part. Deciding how you want to handle NDAs is the work. You can't automate a decision you haven't made.
Start with the checklist already in your head. Turn each item into a test. Then hand it to something that won't get tired of reading the same document for the two-hundredth time.
Parts 2 and 3 are in the Field Guide: the full Wordsmith setup with the playbook template and the register, then the same workflow in Claude, ChatGPT and Copilot.
What's the one NDA provision you change every single time?
