Two Skills + One Automation

Field Note #7

.WHAT YOU'LL LEARN

• How to use AI skills with personality to handle the emails you'd rather not write

• How to stress-test your own documents before they leave your desk

• Why the real AI opportunity for legal isn't better tools, it's better data plumbing

This week I had someone message me asking how to get started with AI at work when their company hasn't approved Claude. My answer was the same as it always is. Start with what you already have. Build something small. Make it useful enough that people notice. An easy way to start is to build a skill file (or if you don’t have skills in your AI, a re-usable prompt), two of which you will find below.

The polite f*ck you email skill

I know. The name. But it's accurate, and that's why it resonates. And in Denmark, kindergartens say “f*ck” so I’ve been desensitized even more!

Every GC has a version of this email in their drafts. The one where a vendor is pushing past the agreed scope. A counterparty is reopening a closed point. Someone internal is asking you to sign something that should never be signed. You know exactly what you need to say. You just need it to not sound like you're saying it through gritted teeth.

The skill does one thing before it drafts anything: it asks whether you want to leave the door open or close it for good. That single question changes the entire output. Ice cold and precise when you're done. Firm but measured when you're not. No passive aggression. No over-explanation. Three to five sentences and a clear close.

I built it because I was tired of spending cognitive energy converting frustration into professional language. That translation work is real work.

The outside counsel review skill

This one has a personality too. A grumpy, battle-hardened senior partner with 35 years of practice and no patience for vague drafting. You upload your agreement, policy or any document before it goes out. It reviews it the way opposing counsel will. It doesn't compliment you on your structure.

What comes back is a tiered issues memo. Critical, Medium, Low. Clause reference, problem, suggested change, next steps. Every ambiguous term. Every undefined tech reference. Every one-sided obligation. Every broken cross-reference.

The point isn't to replace review. The point is to catch what you've miss.

I’ve found when I use soft language like “review this for errors and inconsistencies,” I don’t get as good of results as when I use this skill file. I’m giving the AI more context and in return, it’s giving me a better review.

Before any document leaves your desk, run it through something adversarial. You get better at your own drafting. You send out tighter documents. And you stop finding the problem after the counterparty already has.

Something I'm actually building for sales

I've been having a lot of conversations lately with legal and ops people asking about how to start building with AI. Not which tool to buy. How to actually build something that does something with the tools they have available.

My answer keeps coming back to the same mental model. Think about where your data lives. Then think the piping you need to connect the data to AI. Then think about where the output should land. “WTF are you talking about Laura?” I need an example.

Current State: We use Salesforce. When a deal is ready to move forward, we need to get the data from Salesforce into the Order Form. Right now, a human moves that information from one place to another. Manually. Every time. Gross.

Ideal State: Trigger in Salesforce pushes data to Slack. Slack then gets the data and kicks off a Wordsmith Workflow where the Salesforce data gets dumped into a Wordsmith Blueprint, which is our standard Order Form. Once the data in input into the Order Form Blueprint the completed Order Form is then pushed back to Slack.

In other words, a trigger in Salesforce kicks off an automation. That automation pulls the relevant data and feeds it into Wordsmith. Wordsmith completes the Order Form. Then a Slack message goes back to the salesperson letting them know the form is done.

The data already existed. The task already existed. The only thing missing was the piping between them and something to do the work in the middle (aka the AI).

This is the question I'd ask anyone trying to figure out where AI fits in their function. Not "what can AI do?" but "what data do I have, where is it, and what happens when it can move?"

Legal teams don't usually think this way because legal work feels like judgment work. And a lot of it is. But there's an enormous amount of legal work that is actually just information moving between places with some transformation applied to it. That part is not judgment work. That part is just plumbing. And that work is perfect for AI.

Build the pipes. Then power them with AI.

If you're at a larger organization with an IT department, you're not building this alone. Everyone is trying to figure out how to get the most out of the tools they already have. The legal team often has more data, more process dependency, and more repetitive output generation than any other function.

I'll share how the Salesforce to Wordsmith to Slack automation actually lands when it's live. That's for a future edition. But the framework is the one above. Map your data. Connect the data pools. Apply the LLM where the transformation needs to happen.

Start with what you have.

Happy weekend! Speak soon!

If you have not seen Pink in concert or her documentary, I highly, highly recommend her. My husband works in music, and both he and I were blown away by her last concert - the sheer athleticism is amazing.