So What? Plug-ins for Claude Cowork

Katie Robbert – 00:30

Well, hey everyone. Happy Thursday. Welcome to So What?, the Marketing Analytics and Insights live show. I am Katie Robbert, joined by both Christopher Penn and John Wall. John has internet this week.

John Wall – 00:39

I have internet. This is life on the outer rim. We tend to lose communication now and then.

Katie Robbert – 00:45

Well, we’re glad to have you back this week, John. Two weeks ago on the live stream, we covered Claude Cowork. You can find that episode if you go to the Trust Insights AI YouTube “So What?” playlist.

Last week we covered Claude Code, and this week we’re coming back to Claude Cowork because, in the two weeks since we covered it, things have changed. More stuff has been added, and I think for once I’ve actually used the things that Chris is going to be showing on the live stream. We’ll see.

First question, John—I know you were using some generative AI earlier to work on some responses, but have you done any digging into Claude Cowork, Claude Desktop, or Claude at all?

John Wall – 01:39

No. Like the past two weeks, it’s been all the rage. We covered it in Marketing Over Coffee with Chris, and I know you guys have covered it. So, I’m interested to see all the hooks and plugins this week.

I have not dug in again. I let Chris run around and show off all the cool stuff, and then I’ll jump in and take the cool three things rather than spend any time digging.

Katie Robbert – 01:58

All right, awesome. Chris, I’m nervous. Where do we start?

Christopher Penn – 02:05

All right, let’s do some table settings because there’s some new vocabulary. What has changed in the two weeks since we did a show is that Cowork has added support for plugins. There are four components of vocabulary we need to be aware of. In the Claude ecosystem, there are agents, skills, commands, and hooks.

Commands and skills are pretty much the same thing. They are a shortcut for how to do a specific process. Command is the older version; skill is the newer version. A hook is a trigger. If you think about a trigger in your marketing automation software or your spam filter, a trigger basically says, “Hey, when this happens, do something.” If you use Google Tag Manager triggers, it is a very similar idea.

Agents are little workers—little self-contained things that do a specific thing, like the co-CEO of Katie or the Voice of the Customer. If you bundle together agents, skills, and hooks into a package, that package is called a plugin. In the Claude ecosystem, plugins operate in two places: Claude Cowork and Claude Code. They are cross-compatible, which is important because if you’re using certain plugins in Claude Code and you really like them, you can use them in Claude Cowork with no major additional changes.

The reason why this has caught so much attention is because this adds in all of those skills and agents that we love in Claude Code, but for people who don’t want to fire up a terminal window, it was previously inaccessible. Also, Anthropic bundled 11 of their own Claude plugins for Cowork in a variety of areas with the launch.

Let’s go ahead and take a look. I’m going to shift over to our Cowork environment. Remember, there are three environments. There’s Chat, which is regular conversation; there’s Cowork; and then there’s Claude Code on the web, which we specifically ignored because it’s not particularly good.

In Cowork, there’s a new plugins option down in the lower left-hand menu. If we tap on plugins, you can see the 11 plugins that Claude now comes with that you can install: Bio, Research, Customer Support, Data, Enterprise, Search, Finance, Legal, Marketing, Product Management, Productivity, and Sales. These are the out-of-the-box ones that Anthropic made for users. For fun, let’s install the marketing one.

Katie Robbert – 04:50

I have not used this. I was using the new connectors this morning and got very excited thinking I was getting ahead of the game, but the answer is no, I haven’t used these. This is all still new to me.

Christopher Penn – 05:06

So, now we’ve installed the marketing plugin, and you can see there are now six new commands: Brand Review, Draft Content, Campaign Plan, Email Sequence, Competitive Brief, and Performance Report.

You should tap back on the plugins menu to see what is inside. For the Brand Review command, it shows you the prompt that it uses. This is an important thing for people to know. Please don’t use these and just say, “Anthropic made it, it must be right.” Actually go and look and see what’s in the box.

These are the commands and the connectors that you could install with it if you wanted to. You can see this one has Slack, Canva, Figma, HubSpot, etc. You could install these connectors that would give the marketing plugin the ability to talk to systems.

If you have Ahrefs, HubSpot, and Slack, you can turn on those connectors. Then, as you’re using the marketing plugin, it can talk back and forth to the systems and say, “I want to let the marketing team know that I put the new live stream graphic in Canva and it’s ready for you.” You can see the skills it has: commands, connectors, and skills. That’s what’s in the box.

Now, if I wanted to do any of those tasks, I could say, “Here’s an idea I had. Let’s design an email sequence for it. Walk me through it.” Same rules apply: you still want to give it a folder to work in.

Hey, wait a minute. That’s weird. They revved Opus to 4.6. That was not there an hour ago.

Katie Robbert – 06:56

Well, that’s actually helpful because offline I was going to let you know that I’m getting an error message in my Claude saying the model used in this conversation is no longer available. It tells me to switch to continue chatting, but nothing happens when I try to switch the model. It only lists the two older ones. As of our preparation for the live stream, things were different than they are now.

Christopher Penn – 07:23

Well, here’s what we know behind the scenes. This is just some inside baseball. Anthropic released a bunch of commercials that are going to be airing at the Super Bowl, and it is widely suspected they’re going to release new models either just before or during the Super Bowl as their way of sticking it to OpenAI and ChatGPT. The commercials are actually very funny. They’re on YouTube if you want to go watch them; they’re all making fun of ChatGPT ads.

Back to Claude Cowork. We now have the ability to use these plugins, and you can install multiple plugins in the same way that you can in Claude Code. Let’s browse some more. Let’s install the sales plugins and, Katie, since you like productivity, let’s install the productivity plugins as well.

Sales has Call Summary, Sales Forecast, and Pipeline Review. You could actually connect HubSpot to this and do a pipeline review. The connectors ask for things like Clay, Close, HubSpot, Notion, ZoomInfo, and Fireflies. If you want to do a sales call review, you could say, “Do this and then post the results to Slack.” The skills include account research, call prep, competitive intelligence, asset creation, daily briefings, and drafting outreach.

Finally, if we go to productivity, we have memory management and task management—generic stuff like, “I need to remember how to do things” or “I want to improve my productivity through the day.”

John Wall – 09:07

Is that all out-of-the-box stuff?

Christopher Penn – 09:12

Yeah.

Katie Robbert – 09:15

I have some questions. I know you’re surprised by this. I’m looking at this and thinking, “Oh my gosh, that’s so exciting.” For businesses like ours where we really can’t add headcount at the moment, it’s like adding headcount without having to do it.

But as you’re walking through it, I realize it wants to access all of my HubSpot data, all of my Asana data, and basically my entire ecosystem of how I’m running the business in order to make me more productive. Can we just acknowledge for a second that while that sounds great out of the box, it’s also a little terrifying?

Christopher Penn – 10:04

Yes. You have to trust in the Service Level Agreement and the Terms of Service that you signed when you purchased your Claude account—that it’s as ironclad as it can get with a big tech company.

There’s also a warning. For the legal plugin, they do say this is not a substitute for an actual lawyer. This is more for things like reviewing contracts, triaging NDAs, and prepping for briefings. They explicitly state throughout the tool not to provide legal advice because they don’t want to get sued.

Katie Robbert – 10:46

I look at that and think it could be interesting. John, I know one of the things that can bog down your time is going through NDAs and MSAs sent over by clients to make sure all the language is correct and mutually beneficial.

It sounds like this is something we could use to make that kind of task more efficient, knowing that John—the human who has the most experience and expertise—would still review what Claude Cowork has said regarding where to redline or what language to change based on our standard templates.

Christopher Penn – 11:29

You’ll notice the first part of this prompt is “Playbook Review based on review methodology.” If no playbook is available, it asks the user to make one—check for a completely configured playbook that defines the organization’s standard positions, acceptable ranges, and escalation triggers for each major clause type.

For example, when you’re doing a sales contract, if they say, “We pay Net 120,” and our playbook says Net 30 is the max, this will flag it. But it’s contingent upon us having the playbook. I don’t think we have a legal playbook or a contract playbook yet.

Katie Robbert – 12:04

We have our standard templates, which we would feed in as the playbook. We already know what those terms should be for Trust Insights.

Christopher Penn – 12:14

Yep. Let’s see if there are any other built-ins worth playing with. If you want to do data analysis, customer support, or finance—let’s take a look under the hood at Data Analysis.

This includes things like exploring datasets, building dashboards, creating visualizations, and writing queries to your database. This is where you start to see why plugins are more than just product prompts. If you look here in the Data Context Extractor, there are references like templates and a folder called Scripts that includes actual working Python code to do the data analysis. Claude doesn’t have to try and reinvent the wheel every single time you use the plugin, which is pretty cool.

Andrew Smith says Opus 4.6 has just gone live. Anthropic is in the process of releasing things right now. I remember being at MAICON two years ago when Paul Roetzer was about to get on stage, and GPT-4o was released the moment he was getting on stage. He was like, “Well, there goes my talk.”

Katie Robbert – 13:35

Here is a question for companies like ours, but also larger companies who are using a shared login. We’re all logged into Claude Desktop through our client services account, which is the account that owns the Claude agreement. As you’re installing these plugins on your desktop account, would I see those reflected on mine, or do I then also have to go in and install whatever plugins I want to use?

Christopher Penn – 14:09

I believe you should see them on yours.

Christopher Penn – 14:15

Oh, yeah. 4.6 is just out—literally 13 minutes ago. It scores the highest benchmarks on GPQA. That’s interesting. Wow, it’s really smart. We’ll come back to that later.

Katie Robbert – 14:30

I just refreshed Claude Desktop and I don’t see them. It looks like it wants me to go ahead and install the ones I want. It may be something that gets smarter about those cross-functionalities over time.

Michael Ramos has a good comment: “I think there’s an opportunity to be an agency that builds standards, processes, playbooks, and policies for companies so the guardrails stay intact.” Absolutely. We talk about this a lot. It’s something you should have before you even touch the technology.

I believe the thing we would want to call up is the 5P Framework: Purpose, People, Process, Platform, and Performance. Don’t touch platforms until you have the first three. Just for those who are curious, we do that kind of work. We help you put together all of those playbooks and guardrails. You can always reach out to John Wall at TrustInsights.ai/contact. But that was a really good reminder. Thank you.

Christopher Penn – 15:41

Okay, so those are the plugins, and you can see they can run code. For plugins, you can add them by Anthropic, from GitHub, from a URL, or by uploading a plugin. Please do not install plugins from people you do not trust because they can run hostile code in your Claude Cowork.

If you’ve given it rights to things like sensitive data and you don’t know what that plugin does, don’t do it. You can inspect what’s in the plugin and you should. Particularly if you’re going to install a third-party plugin that’s not from Anthropic, please take the time to look through it. If you don’t know code, ask Claude to inspect the plugin and ask if it shares data externally or if there is any telemetry built in. These are standard security questions.

Katie Robbert – 16:38

You’re saying that third-party plugins can inject hostile code, but there are also just a lot of shiny object chasers out there who might have positioned their plugin really well but don’t necessarily have the correct security protocols or data privacy. So, if you see anything in GitHub built by myself or John Wall, definitely don’t install it because neither of us knows what we’re doing. Chris knows what he’s doing; John and I, not so much.

Christopher Penn – 17:21

The other thing is, as with all AI stuff, if you’ve got stuff and you don’t need it, you can see there’s a little button to turn it off. Turn off the plugins that you’re not using because when Claude initializes, it does consume a little bit of the context window to know those things are there.

It doesn’t consume a ton, and the new version of Opus apparently has a one million token context window, which is a big change. However, it’s still a good practice to turn off stuff you don’t need. If you’re doing sales work and you have the legal plugin installed, just turn it off. Claude in particular is very smart about listening to what you’re saying.

Christopher Penn – 18:03

If you’re saying, “Let’s do some sales stuff and I need to look at the sales contract,” and you have both legal and sales installed, it might say, “I’m going to do a legal contract.” If you’re like, “No, I only wanted to work on sales contract review,” turn off the things you don’t need.

If you really don’t need it, just uninstall it and remove it from the upper right-hand menu. It doesn’t damage Claude in any way, and for the Anthropic ones, if you decide you need it later, just go back and install it again. There’s no harm. I’ve uninstalled and installed many times just to see if I could break it.

Katie Robbert – 18:34

Why not? But I think that’s really good. One of the things, Chris, that you say a lot about generative AI is that it’s a really over-eager, helpful intern. If it has all of these different things to pull from, it’s going to.

That is why we highly recommend getting yourself organized with the 5Ps and why we recommend projects all have user stories to help you focus. These tools are expensive, they go off the rails, and you want to make sure you have a clear vision of what you’re doing going into it.

A question we got earlier was how to feel okay about the economic and environmental impact of using AI. My advice, knowing this person is not a technical user, is to be really thoughtful about how you’re using it. Don’t outsource your thinking to it. Really think about why you’re using it and what you want to get out of it. If you’re just asking random questions because you’re curious, that’s when that environmental impact starts to build up. But that’s probably a topic for a different show.

Christopher Penn – 19:59

No, it’s definitely a topic, but an important one and worth doing. All right, let’s turn the marketing one back on. Let’s see what this looks like so we can get a sense of it. I’m going to say, “Let’s do a campaign plan review.”

When you start typing in the slash command, you’ll notice that you have both the plugin name and the skill name. You want the shorter one, “Campaign Plan” in this case. I’m going to drag and drop in the Inbox Insights growth strategy, which we did a while ago on how we get our newsletter to grow. We’re just going to say, “Just review this. Tell me what you think about this.”

Christopher Penn – 21:19

Let’s hope that Anthropic’s APIs are working at all, having just launched a new model. In Cowork, it’s going to start by loading up the skill to understand it, taking the pieces apart, doing its data extraction, and running its commands.

Oh, it’s going to have questions for us, Katie. What would you like the focus of this campaign review to be? Full Strategic Audit, Feasibility Check, Gap Analysis, or Actionability Review?

Katie Robbert – 21:29

Let’s do a Full Strategic Audit.

Christopher Penn – 21:31

Full Strategic Audit. What format do you want? A Word doc, a chat summary, a spreadsheet, or something else?

Katie Robbert – 21:37

Word doc is fine.

Christopher Penn – 21:38

Word doc. All right, a full strategic audit delivered as a Word document. It now has to load its DocX skill as well to get that out.

Katie Robbert – 21:50

I’ve been using Cowork for quite a few things, and it is always interesting when it goes through and—if you don’t give it direction—it asks, “Wait, what is it you want me to do with this? Here are some options.”

I really appreciate that it asks those questions because it makes me, the end user, think about what I am doing with this. Having those questions up front about whether this is a full strategic audit or just feasibility is really helpful because those are two very different asks.

Christopher Penn – 22:27

Exactly. It’s doing a bunch of research on our strategy plan to validate if we are even strategizing about the right things. Apparently, one of the new things in this model is that it’s really good at knowing what sub-agents and tasks to autonomously come up with. In previous versions, you would have to say, “Use your web search tools to do this thing.”

Now for some reason, it’s building its own JavaScript. That might take a while.

Katie Robbert – 23:25

In the meantime, Chris, we got a question. Let’s assume marketing needs to be in legal compliance. Does turning off a legal plugin remove the safe harbor guardrail, or are we still working from the RAG to stay in compliance?

Christopher Penn – 23:47

It’s really interesting. Anthropic doesn’t use a RAG. Boris Cherny on the Claude Code team talked about this. They found that RAG, which stands for Retrieval-Augmented Generation, delivers substandard performance compared to good old-fashioned search. So, in all incarnations of Claude, it just does straight-up search.

In terms of Safe Harbor guardrails, it depends on your Claude subscription and what you are paying for and where your data is being processed. You can, for additional fees, pay for data processing in a certain section.

I have to put up the required banner here: we are not lawyers, so please consult with your legal team about how Safe Harbor applies to you and your company. But this is being processed by Anthropic, presumably in the nearest data center just for cost reasons.

Katie Robbert – 24:50

John, speaking on legal—though we would never use this for legal advice—would you consider using a system like this to do the first pass of an NDA or an MSA that’s come across from a potential client to see where you need to redline? Or is that something you would still want to do yourself?

John Wall – 25:14

My take on that has changed. At first, I was like, “No, it’s not worth doing because I still have to do the full comb-through.” But I’ve realized what it needs to do is exactly what we were talking about with our playbook.

I need to put together three or four pages of the things I look for. I have that process baked in; when I come to a section, I know the four gotchas to look for. If I just made a document of all those things, then I could say, “Here’s the Trust Insights patented NDA examiner skill.”

The great thing is that it could easily go through and say, “Okay, none of these problems are in these five sections, so you only have to read these two sections.” That could save a ton of time. I am confident in doing it. The big thing is to make sure you’re using the process of it being able to categorize to get all the stuff that’s okay out. If I can have it say, “You don’t need to read these three of the 64 sections,” that can save a ton of time.

Katie Robbert – 26:43

I feel like I need animation for these moments because, folks, we just had a breakthrough live on air. We had our laggard on the innovation curve—John’s on board! John’s on board with using AI!

I need us to just acknowledge this moment. It’s not that you don’t believe in it—you obviously work with us—but for your own personal use, this is such a big deal and I love the way you’re thinking about it. John, you’re a good representation of a lot of industry professionals who feel it’s just easier to do it themselves. Seeing that switch flip where you realize if you just do this, it’s going to save you so much time, that’s such a huge moment.

John Wall – 27:36

Yeah, well, it still comes down to the truth that I have to do it the first time and confirm that it actually works. If it still comes back and says I need to read 27 of 27, we’re going nowhere.

The other thing that can’t be downplayed with the stuff I’ve been doing recently is that I’m building these prompts based on the 65 skills that you and Chris have already put together. I just say, “Hey, obey the rules that these guys have spent two years putting together.” Now I look really smart, but the fact is you guys have done a ton of work to get me there.

Katie Robbert – 28:14

We all rise together, John.

John Wall – 28:16

That’s true. This is why you need to call Trust Insights. You don’t want to be doing this yourself. Have us show up with our wagon full of 2,000 prompts that we’ve been beating on for three years and save yourself literally hundreds of thousands of dollars.

Christopher Penn – 28:37

“Hire our wagon of prompts”—that’s going to be our new tagline.

All right, it finished its analysis. This is a very long, multi-page audit of our newsletter strategy. The goal is to get to 40,000 subscribers. It went through section by section and said: Campaign Overview, strong; Target Audience, needs work; Key Messages, needs work; Channel Strategy, adequate—that’s damning with faint praise; Content Calendars, adequate; Content Pieces Needed, strong; Success Metrics, adequate; Budget Allocation, needs work; Risks and Mitigations, adequate; Next Steps, strong.

In each of the sections, it tells you its reasoning, what the strategy does well, and what needs improvement. You can then take this and—the logical next step would be—”Okay, Claude, you have my original and your audit. Write me a better one.” I’ll come back in 15 minutes when you’re done.

Those are the plugins that are built-in. Now here’s where things can get fun. In Claude Code, there’s a plugin development plugin. Last night, I decided I wanted to build a plugin to help me write a press release. I wanted to write a press release for AI, not for humans.

We know that AI sees things differently and, if you’ve taken our generative AI use cases course, you know that if you’re building press releases for AI, you’re going to do it in a different way. To start, you need to have some idea of what AI is looking for. We started with our AI Readiness Course, which is the Brand AI Readiness Assessment—a brand new thing Katie came up with.

I said I want a plugin to help me write this press release calibrated against what generative AI sees. My first step was to go into our private GEO tool and ask what some good AI readiness assessments are. It exported this massive amount of data that I then stored as a file.

I built this plugin because I want to semantically match press releases to what Google is doing in AI Overviews. Google has specific words, phrases, and concepts it is expecting. I need to know the technical evaluation criteria and scoring rubrics in the Microsoft, Cisco, and AWS AI readiness assessment tools to benchmark organizations.

If I want my press release to resonate semantically and topically with what Gemini is asking itself, I need to use those concepts. In the plugin, there is the skill and then a piece of Python code I wrote that does the text metrics. I gave it two paragraphs of me “foaming at the mouth” about the new assessment and copied the landing page content.

Then I took the Gemini trace of what Gemini was asking itself as part of AI Overviews and said, “Claude, use the plugin.” The plugin does two drafts of a press release. It says these are the terms that need to be in here: AI Readiness Assessment, AI Project Initiative, Governance, ROI Framework, Maturity. These are the “nice to haves”: Strategy, Leadership, Executive. It reads that and spits out two press releases, then scores them.

It said Candidate B, the second process, scored highest. I said, “Great, now make a third one and try to get your score even higher.” It came up with one and got a 48 out of 100, which is actually pretty good. What went out on the wire last night was this gigantic press release that, unlike previous GEO press releases which were horrendous mangled piles of words, actually reads coherently.

Katie – 34:56

That was the question I was going to ask: who is the audience of this press release? Previously, the audience was the machine, not the human. Now it sounds like the audience is both.

Christopher Penn – 35:25

It’s still the machine.

Katie Robbert – 35:26

Still the machine, which I think is an important distinction. We are not in the business of public relations, but we are in the business of helping people understand how to show up in generative engine optimization. When someone puts a chat into Gemini, they want to know how often they’re going to show up. This is what this does.

Christopher Penn – 35:58

Exactly. This semantically checks the boxes on how Google’s Gemini sees language and understands what it means without making it just a big bag of words. Even the quote in here sounds like Katie: “Every week we talk to organizations that have invested heavily in AI tools but haven’t addressed the fundamentals. Their data quality is poor, the data hygiene and data availability practices are inconsistent, their processes aren’t documented, and their team lacks the AI literacy skills to manage and audit AI outputs. This AI readiness assessment helps surface those organizational readiness issues before they become expensive project failures.”

You can now in Claude Code make a plugin that is just a zip file of the prompts and scripts. You hand that to your team members and say, “Install this in your Claude Cowork,” and you’re up and running. It is the fastest, easiest way to add new capabilities.

I don’t see much in operations management or strategy in the standard set of 11 plugins. I don’t see anything on public relations. Your organization could make versions of this for yourselves. You might even make them and sell them to other people.

Our friend Gini Dietrich, for example, is the owner of the PESO model. She could create a PESO plugin for Claude Cowork that could audit your public relations strategy. Any PR manager pressed for time can say, “Here’s our 2026 strategy, PESO plugin, go.” Because these skills can connect to systems, it could ask, “Do you want me to come up with a project plan and put it in your Asana to get you aligned to the PESO methodology?”

Katie Robbert – 38:59

I feel like there’s a lot of potential with these plugins, especially for companies looking to create their own. But again, the caveat is that it is software development. You need to follow the SDLC, which is the Software Development Life Cycle. AI has not made that go away; if anything, it has highlighted the need for structure and guardrails. Just because AI can do it doesn’t mean you should just let it go. You still have to have a plan and requirements.

Christopher Penn – 39:57

Yep. And if you’re using Cowork, it’s an agentic environment, which means that it can take your mistakes and make them faster. It will amplify them.

That’s the big difference between this and GPTs. GPTs are not autonomous; you are still doing all the manual work of copy-pasting. With systems like the new version of Claude Cowork or the new versions of Google Astra, these are agent systems. Instead of you having to do all the typing, if you provide the materials up front, you set them off and say, “Go,” and you come back when the work product is done.

Christopher Penn – 41:15

Michael pointed out that it doesn’t always do the best work the first time. The reason for that is because it’s a probabilistic tool. You give it a starting point, and if you are specific—or your plugin is specific—about what “better” means, then it knows how to make it better. In our public relations plugin, there’s a Python script that does semantic indexing. It gives Claude a goalpost so it can score versions and keep trying. You have to be clear or have a mechanism in your plugin to say, “This is what better means.”

John Wall – 42:05

I had to laugh when you ended with the power and stability of the platform, and we opened the show with getting the engine ripped out of the plane and replaced as we’re taking off. Process is very important here to avoid a dramatic malfunction.

Christopher Penn – 42:32

So, Katie, what plugins are you going to have me build? I can very clearly see a co-CEO plugin getting installed in Claude Cowork.

Katie Robbert – 42:46

I think we’re definitely going to do that. The thing we’ve been getting a lot of asks for is AI alignment. The foundation would be the 5Ps. I could see building something for someone to interact with who has taken the AI Readiness Assessment and doesn’t know what to do next.

The majority of the issues are not around having the right AI tool; it’s around people and process. We were talking with a client earlier and asked if there was a ticket for requests. They said no—sometimes people come by the desk or send an email. That’s the problem. It’s not the platform; it’s people and process.

Christopher Penn – 44:38

I definitely see “Co-CEO” coming and “Ask Our ICP” coming to Claude Cowork. You could say, “Here’s a campaign plan. Co-CEO, audit it. ICP, tell us what you do and don’t like about it.”

Katie Robbert – 44:55

And once we have our legal playbook, I think that plugin would be helpful to take some of the burden off of John so he can be that last set of eyes to see what the machine didn’t catch.

Christopher Penn – 45:20

Finance, law, and health are the magic three. Never use AI outputs in those areas without a human expert reviewing them. The trashy romance novel doesn’t have the same legal risk as the contract you’re about to sign for $100,000.

John Wall – 45:43

That’s the “Your Money or Your Life” data.

Christopher Penn – 45:47

Exactly right. All right, so that’s going to do it for this week’s episode on Claude Cowork plugins.

It’s going to be very busy the next two weeks because we’re coming up on Lunar New Year. In the AI world, that means model makers in China, Japan, and Korea are racing to get new technologies to market. This puts pressure on Western companies like OpenAI, Anthropic, and Google to also come out with announcements.

There’s going to be a lot of noise. Drop by our Analytics for Marketers Slack group if you have questions about what an announcement means. Make sure you’re subscribed to the Trust Insights newsletter and we will talk to you all next time. Thanks for watching.