Claude Co-work for Lawyers: What Works (Beginner’s Guide)

I've been running Claude Co-work across legal workflows for about the past 2 months. Contract reviews, NDA triage,

months. Contract reviews, NDA triage, custom bill automations, compliance checks, regulatory monitoring, and I've been logging what's working well and what's not. And so, what I'm going to

what's not. And so, what I'm going to show you in this video is what Claude Co-work actually does really well for legal teams. I'm going to tell you what it is and how it's different from just the normal Claude. I'm going to show you

where it falls short and a few settings that you need to change before you put a single client document anywhere near it.

There's a lot of what I would call power user tips and maybe I'll save them for another video, but this video is just going to be a nice easy introduction into what Claude Co-work is and how you can leverage it. And later in the video,

I'll show you how you can get a free ROI calculator that I built actually with Claude Co-work so you can plug in your firm's actual numbers and see if Co-work is worth the investment for your team.

Before we touch on anything else, if you're on the pro or max plan and so you're not on the enterprise plan, your conversations are used to train anthropics models by default. And this

includes everything that you upload to co-work. And here's me being Captain

co-work. And here's me being Captain Obvious, but for a law firm handling client contracts, NDAs, confidential documents, that's obviously a problem.

So, I'll just quickly show you how to turn it off. Just come down to your profile, click on settings, come down to privacy, scroll down where it says help improve claude, just make sure that that's toggled off. And that's all you need to do. And if you're a firm or a

legal team that handles anything regulated, you probably want to get the enterprise plan because there's absolutely no chance that your data is used for training that's in the contract. That's a contractual

contract. That's a contractual obligation from Anthropic. And so it's just a safeguard that I think is worth the investment. Okay. So what is Claude

the investment. Okay. So what is Claude Co-work? Now, quick context if you

Co-work? Now, quick context if you haven't used it. Co-work is like a legal assistant. It's built into the Claude

assistant. It's built into the Claude desktop app. So, not the browser app.

desktop app. So, not the browser app.

You give it access to a folder on your computer. So, you can choose specific

computer. So, you can choose specific folders. Definitely don't give it access

folders. Definitely don't give it access to all your folders, but then describe what you want done. And Claude goes and does it. And so, as opposed to where a

does it. And so, as opposed to where a typical chat feature in any of the LLMs responds to the question or the prompt that it's being given, think about using co-work as a means to getting you to a

specific outcome. So, it reads your

specific outcome. So, it reads your files, creates files, edits documents, it connects to applications and tools that you use, and then it runs multi-step tasks without you having to

prompt it every 30 seconds. I.e., it's

getting to the outcome that you want to get to. Or to put it another way, think

get to. Or to put it another way, think of working with Claude Co-work, how you might work with a junior associate. You

might say, "Here's the playbook. Here's

the folder. Go." And compared to most legal AI tools on the market, Co-Work is one of the few that actually works directly with local files instead of forcing everything through the web and an upload. And that's really powerful,

an upload. And that's really powerful, right? Cuz you can actually have it

right? Cuz you can actually have it doing tasks on your folder as opposed to just searching and giving you information for you to do the task. It's

available on Pro, Max, Team, and Enterprise Plan. The Mac version, the

Enterprise Plan. The Mac version, the Mac app is really solid. The Windows

version is still sort of a bit buggy and there's a lot of complaints, but I'm sure they're going to improve that soon.

So, like I said, it works with local folders. So, you want to identify what

folders. So, you want to identify what folders on your desktop or your machine that you are going to work inside of.

It's always a good idea to sync them with some sort of cloud storage. So,

SharePoint or Google Drive or whatever it may be. And maybe make duplicates as well if it's valuable important things that you're going to have Claude Co-work

on. I would absolutely recommend create

on. I would absolutely recommend create duplicates because sometimes AI goes awall and it can do things that are very much and you don't want to be in that situation where you're losing valuable

work and it's changed it and there's no revisionist history. So, it's just

revisionist history. So, it's just better to be safe and have a duplicate version. So, let's get into it. Let's

version. So, let's get into it. Let's

select a folder. Here we go here. So,

I'm going to go to my desktop. All

right. And you'll see there's a warning.

Are you giving it permission to edit the files in this folder? I'm going to click allow.

Okay, now that I've selected the folders that I want to work inside of, which actually have some contracts inside of them, I'm just going to come over here and I'm going to select slash review.

And you'll see here there's an option review contract. And you might be

review contract. And you might be wondering, okay, where's this coming from? So, this is actually a pre-built

from? So, this is actually a pre-built installed skill from the Claude Legal plug-in. If you're not sure what that

plug-in. If you're not sure what that is, I've made a video about it. I'll

leave a link in the description to that video below. But essentially what it is

video below. But essentially what it is is a series of instructions for common legal workflows. They are fully

legal workflows. They are fully customizable so you can tweak them and make them more bespoke to your firm or your legal team. So you can see here it's actually giving me an option how I want Claude to review the contract. So I

could absolutely upload it. But because

we've connected our folder, we're just going to select the second option, right? It's going to reference a

right? It's going to reference a contract file from our system. So we'll

click that. We'll click go. So you can see here it's querying. It's looking

what's inside of my folders. says it. It

says, "I see two contracts in your workspace, a sample NDA and a sample SAS vendor agreement. Let's have it review

vendor agreement. Let's have it review the sample SAS vendor agreement." And

it's going to ask which side of the deal that we're on. We are the buyer and any specific focus areas or deal context that I should know about. Uh IP

ownership is important, liability, indemnity, and a general review. Now,

just look at the process that it's going through. And obviously, let's look at

through. And obviously, let's look at the output when it's ready. But you're

going to see a much more detailed and useful review of a contract than you would just typing review this contract or summarize this contract and compare it to an existing contract because it's got the instructions inside of that

skill, which I've gone through all of them now, and they're all really good.

They're a great foundation for which you can build upon and customize and edit to your own firm standards. So, while

that's working, just take a look at the right hand side of the screen here. So,

you can see that firstly, it developed a plan, right? Right. And so it's now on

plan, right? Right. And so it's now on step four of five. Uh you can determine which folder it's working out of, specifically which file it's working out of. And then you've got the skills that

of. And then you've got the skills that it's using to analyze the contract or perform the given task. Okay. And now

that has finished. And so as you can see here, it doesn't just give you or it doesn't just flag things. It gives you specific redline language that has a justification, right? So, it's got

justification, right? So, it's got suggestions uh replace this clause with that uh along with a rationale that you can actually use uh presumably in negotiations. At least that's what

negotiations. At least that's what lawyers have told me. Now, is it perfect? Well, obviously not. You still

perfect? Well, obviously not. You still

need a lawyer reviewing every output, but no one should be using AI to replace lawyers judgment. But as a first pass,

lawyers judgment. But as a first pass, that catches the obvious stuff and surfaces the things worth perhaps arguing about. It saves hours, that's

arguing about. It saves hours, that's for sure. That first red line that used

for sure. That first red line that used to take an associate maybe 3 to six hours. This gets you 80% of the way

hours. This gets you 80% of the way there in just a matter of minutes. Next

thing we're going to show you is how to do NDA triage again using the Claude legal plugin. I'll quickly just show you

legal plugin. I'll quickly just show you how to install the plugin. So just come down here, go to plugins, and then there's even a a default category. And

so you can see all the different pre-baked skills that are there ready for lawyers to use out of the box.

They're all customizable. Again, that

video is in the description showing you how to do that. So, as it pertains to NDA triage, if your team processes dozens of NDAs a month, this is where co-work earns its keep in my opinion,

the /triage NDA command categorizes each NDA as green, yellow, or red. So, I'm

just going to type that in. Triage NDA.

And you see, I don't even need to press anything. It already knows what I want

anything. It already knows what I want to do. And so, it says, how do you want

to do. And so, it says, how do you want to submit the NDA for triage? We are

going to just skip this. Let it just read our folders. and let's see if it can find

folders. and let's see if it can find it. It says it's found the NDA triage

it. It says it's found the NDA triage workflow. That's the skill through the

workflow. That's the skill through the claude legal plugin. Let me check for any firm specific NDA playbook in your workspace and then we'll get started.

And so then it said good. I've loaded

the firm's NDA playbook. Okay, great.

Now that's finished. So green means standard terms, route it for signature.

Yellow means specific issues that need a quick review. And red means I guess sit

quick review. And red means I guess sit down and read the whole thing. for a

team drowning in routine NDAs. This

alone can save a parallegal an entire day per week, I would imagine. And then

finally, just the last thing with the legal plug-in because I've already done a video as I said, but there's a cool feature that I didn't show last time.

So, I'm just going to type in slash brief right here. What briefing do you need? Daily summary, specific research,

need? Daily summary, specific research, or incident response? And so, you can see I've got daily summary of relevant legal departments, research on specific commercial dispute topic, incident response briefing. Let's just do a daily

response briefing. Let's just do a daily summary. Let's go. So, at the start of

summary. Let's go. So, at the start of your morning, co-work pulls together your pending contract requests, upcoming deadlines, and anything that needs attention. It's not magic. It's just

attention. It's not magic. It's just

organized context. And that's exactly what most lawyers are missing at 8 a.m.

So, you can see it's it asked to connect to a bunch of tools that I commonly use.

And then it's going to give me my daily brief. I won't go through it because

brief. I won't go through it because it's going to be my daily brief and it's going to be relevant to you, but you get the idea. Okay. And so, the next thing I

the idea. Okay. And so, the next thing I want to talk about is in fact skills.

So, just to be clear, the Claude legal plugin is a combination of skills, but you can build your own skills. You can

edit existing skills. Now, word of warning. If you take skills from a

warning. If you take skills from a public database or a public repository, run them through another LLM separately and ask the LLM to scan them for any

security vulnerabilities. People are

security vulnerabilities. People are inserting malware and prompt injection attacks into these skills. So, you got to be careful, but you can always create

them yourselves. And in fact, I strongly

them yourselves. And in fact, I strongly urge you to do that because they're going to be built to your specific needs, your specific workflows, and get you the output that you actually need.

And so, I won't go into too much detail.

There's many different ways you can do them, but just a simple way to do it is to have Claude interview you. So, just

tell Claude what you want to do. And I'm

just going to use a voice transcription here. And I'm going to say, "Uh, okay.

here. And I'm going to say, "Uh, okay.

So, I want you to build out a skill for me. We are a real estate law firm, and I

me. We are a real estate law firm, and I want the skill to be about, uh, payment application processing." Essentially, I

application processing." Essentially, I want you to verify the amounts. I want

you to I want you to check the amounts against contract values. I want you to confirm completion percentages and compile master applications. And so,

just build that out. Ask me any questions that you need to get further details from me and then package that as a skill. And so, I've just transcribed

a skill. And so, I've just transcribed that. I'm going to send that to Claude

that. I'm going to send that to Claude and then it's going to come back and it's going to really build it out. It's

going to ask me things that I haven't thought about putting in there. And

something that's great about co-work is there's actually a skill to create skills. So meta, I know, but it's

skills. So meta, I know, but it's basically instructions to Claude to build out really effective skills. And

so you can see that I didn't even ask for Claude to use that. But that's the other thing with skills. You can

manually uh invoke them by using a slash command, you can ask, you can type, you can voice command to use a specific skill, or if you're just asking for something that Claude determines is

going to benefit from using that skill, it will naturally invoke that skill. And

so you can see here it's starting to ask me questions, right? So what format are you typically working with for payment applications? I'm not going to I'm just

applications? I'm not going to I'm just going to randomly select some questions here. And then Claude's going to go away

here. And then Claude's going to go away build a skill. And then you can see it says I have a few more questions to make sure the verification logic covers what you need. What specific checks matter

you need. What specific checks matter most when verifying amounts? Select all

that apply. Over billilling and change order tracking. Let's skip through.

order tracking. Let's skip through.

Let's check a couple more.

And so if it's not already clear, the real value in a skill is that you don't need to keep reexplaining your firm standards every session or your preferred ways of working. You write it

once and Claude follows it every time.

And here's why that matters. When a new associate joins, they don't get a binder of policies. They inherit the team's

of policies. They inherit the team's accumulated knowledge through these skills. It's institutional memory that

skills. It's institutional memory that actually persists. Then you can actually

actually persists. Then you can actually share skills very easily. You can

obviously do it in the manual way where you can download them and send them as a MD file which is just like a PDF or a DOCX. If you have an enterprise plan,

DOCX. If you have an enterprise plan, you can have an admin that actually shares them with all employees or some employees. So there's a lot of control

employees. So there's a lot of control that anthropic has thought about. Now

I've really just scratched the surface of what's possible with Claude Co-work, but that was the intention. I'm going to record a video in the next few weeks that shows you some really advanced use cases. not difficult by any means, but

cases. not difficult by any means, but they're just things that you learn over time. How to leverage the context memory

time. How to leverage the context memory effectively, how to transfer memories uh from one session to another. And that's

in fact one of the shortcomings of Claude Co-work. There is no memory

Claude Co-work. There is no memory between sessions. So, every co-work

between sessions. So, every co-work session starts fresh. It won't remember what you did yesterday. So, you need to make sure that you're creating skills and memory files and logs that keep

track of what you did in each session and share that with the next session.

And that's what I'll show in the in an upcoming video. And in case it's not

upcoming video. And in case it's not super clear, when we say session, we just mean like this is a session. Each

one of these is a session. When the

context memory of this fills up, uh even if I'm on the same task, I'm going to need to start a new session. Co-work

does burn through tokens fast. If you're

on one of the larger plans, it's not going to be such a concern. Co-work does

use significantly more than the standard chat because it's running multi-step tasks. It's reading files. It's running

tasks. It's reading files. It's running

searches. and it's coordinating sub agents. If you're on pro, you definitely

agents. If you're on pro, you definitely will hit the limits pretty quickly. So,

just monitor your usage and settings if you are on pro. Thirdly, if you are on Windows, then I would say as of March 2026, proceed with caution because the desktop app has known issues.

Installation failures on Windows 11.

It's crashing that require full reboots and a background virtual machine that causes mouse stutter even when you're not using co-work. That's how Chord Co works on your machine. They both have a

mini virtual machine. So the Mac version is there. It's stable. The Windows

is there. It's stable. The Windows

version is not quite there, but it won't be long. So certainly don't let that put

be long. So certainly don't let that put you off using it, or at least don't let it stop you from using it permanently because the benefits certainly outweigh the negatives. And fourth, and this is

the negatives. And fourth, and this is an important one for regulated firms, co-work activity doesn't show up in anthropics audit logs, compliance API,

or data exports. If you need an audit trail of what your team is doing with AI, that's a gap, honestly. right now.

And fifth, the default playbook is US focused. Delaware, New York, California.

focused. Delaware, New York, California.

If your firm operates in the UK, EU, or Australia, or anywhere else, you certainly need to customize before you rely on any of the contract review outputs. And I know I briefly already

outputs. And I know I briefly already touched on this, but I just want to mention one more thing about how to create and set up your folder. Don't

point at your entire documents. As I've

already said, give it a specific workspace with subfolders for contracts, templates, and outputs. And number two, always use context files. So drop a file called Claude MD into your workspace

folder with your firm's preferences, jurisdictions, standard terminology, formatting requirements. Claude reads

formatting requirements. Claude reads this automatically every session. I'll

show you what I've written here just to give you an idea. You got you can see we've got firm identity, rules, style and formatting, reference materials, document templates, current matters.

Claude knows to read this every session.

And so this is really important information that you want to include that's carried across multiple sessions.

Number three, you want to describe the end result, your objective, what you want to get to out of the session, not the steps, per se. So don't say things like, "Open the file, read paragraph 3,

compare it to clause 7." Instead, just say, "Review this agreement against our playbook, and flag anything outside of our standard positions, and then just let co-work figure it out." Now, it's going to ask you questions, but

eventually it will get there. Number

four, do batch related tasks. And this

is really just about working efficiently from both a time standpoint, but also a contextual memory and token usage standpoint. So if you're reviewing three

standpoint. So if you're reviewing three NDAs, do them in one session instead of three. It saves tokens and it keeps the

three. It saves tokens and it keeps the context consistent. Number five, always

context consistent. Number five, always request a change log and a status MD.

Now, what you can do is build these into your skill files. So, when you're having Claude interview you and asking you what you want the skill to achieve, make sure you ask for a change log and a status MD. What the change log is going to do.

MD. What the change log is going to do.

It's going to give you a report of what changed at the start of every session that happened in the previous session or you can have it tell you at the end of each session. And then the status MD is

each session. And then the status MD is essentially the solve for the fact that right now there's no memory between sessions. Especially important if you're

sessions. Especially important if you're on an enterprise plan because if you have multiple attorneys working on the same case file then creating a status MD you're instructing Claude co-work to

update what was done in a given session at the end of a session and then when a new session is started you instruct Claude to read those notes so that the next session regardless of whether it's

another attorney that context that history of what was done is carried over. And finally, this might seem

over. And finally, this might seem counterintuitive, but if Claude or any AI starts to go a little bit awol, it's always better just to start a new session. Don't get into the habit of

session. Don't get into the habit of trying to fix it, getting mad, prompting it. You're going to spend more time.

it. You're going to spend more time.

It's a sunk cost fallacy. Just move on to a new session. Go back, copy the prompts that you liked from the previous session, and then just start a fresh. It

just muddles the memory when you try and improve an LLM or an AI agent when it's gone down the wrong path because it also maintains the wrong thinking and that wrong path for the remainder of that session. So, it's always better just to

session. So, it's always better just to start a new session. So, [clears throat] here's the bottom line and look, not every law firm is structured in such a way where they're focused on billable hours, but most are, right? And so,

lawyers on average bill about 3 hours out of every eight hour day. The other

five go to admin, formatting, research prep, and all the other things that don't generate revenue. Co-work can fix that. And honestly, it can have a really

that. And honestly, it can have a really meaningful impact. Roll it out, get

meaningful impact. Roll it out, get attorneys starting to use it. Start with

the skills that I'm talking about today and then advance to full workflows.

They're easier than you think. And in

the next video or a few videos, I'm definitely going to show you how to build out entire workflows for very specific processes inside of law firms. Now, I built a free ROI calculator that

lets you plug in your firm's numbers and see exactly what co-work could save you.

The link is in the description. It's at

legal ai.com.

There's a whole host of other tools that we've built that are focused on helping law firms and legal teams evaluate the effectiveness and the efficacy and the ROI of adopting AI through a range of

different scenarios. So, check it out if

different scenarios. So, check it out if that's of interest. And if you found this useful, please do drop a comment.

I'd be very grateful. And uh if you want to see a new workflow or a specific use case of Claude co-work, let me know and I'll think about making one in the future. And if you haven't already, the

future. And if you haven't already, the Claw Gemini and Chat GPT and Perplexity guides are all linked in the description below. Thank you so much for watching

below. Thank you so much for watching and have a great