Why Claude Tag is Dangerous

So Neil, did you see that Claude Tag came out? Here's what Claude Tag does, right? So let me just show you on my screen over here. So basically these are like, I've been working like this one inside of Slack for a while, right? So imagine that you can talk to your data inside of Claude. So this, they're like, oh, it's a new way for teams to work with Claude. So you can have Claude and Slack team up together and you can combine it with HubSpot, you can combine it with Asana. So you can like a lot of the things we've been doing already, right? But a lot of people are already using Claude. And so this video got 15.7 million views on it. And you can see here, like this person's asking Claude questions. Hey, what should we do for our cadence picker for how often we talk about this product launch? Or over here, it's like, hey, here's our scheduled export for this launch. This is in beta. Would you like me to add this and this thing over here? So your team starts to work inside of Slack and it's with your agents. And it becomes multiplayer. Now, Neil, let me get your reaction on this. How cool do you think this is? Are you guys going to use this?

I think it's cool. We definitely would end up using it.

Okay. I think it's dangerous to use this. Let me tell you why, Neil. So Claude Tag is actually a Trojan horse, not because Anthropic is trying to be evil, but think about it this way, right? So all the things that you have—

They're just locking you in to just make money.

What's that?

They're locking you in to make more money. That's really what it is.

So what they're locking you into is it's not just— we're not talking about model lock-in where they're forcing you to use their models, but it's context lock-in, right? So Neil, if you start to compound all your knowledge inside of Claude and that, it hangs there and like people quit your company, it's not like you can share that memory across, but that entire Claude system has your memory as a company, right? And so you're actually better off making your own or going with like a managed service here where someone puts it together for you. But if you want to make your own, Like, then you can choose what models, you can choose token routing, you can choose token optimization, but you can also choose the scaffolding that you want with the memory, the storage, permissioning, governance, all these things. Yeah, I think the other thing too is just so everyone knows that Claude Tag, it doesn't charge you on a per-seat model like a typical Claude Teams or enterprise subscription does. It actually charges you on usage. So you are paying on basically API. And so I think again, it's going to be extremely useful. I think, like, remember what I said, Neil? I was like, when I did my talk a couple months ago, I was like, most companies are going to be working like this. I didn't anticipate that they would do this as quickly. I thought they would do something like this eventually, but it's faster than I thought. But I think, sure, you can export and things like that, but do you want to be paying on usage? And if you're going to do this, like, there's limited connectors right now. You can add more, obviously, but I also don't want Based on the permissioning in there, like, you can have different people that are in your Slack that maybe they aren't Claude users, they can get access, they can ask for things, right? Then you start to think about prompt injection, you start to think about other threats that maybe those people aren't thinking about. So there's a lot of lockdown that needs to happen here, again, from a permissioning, governance, security standpoint. I think Claude's going to do this. I think OpenAI is going to do this. I just think it's dangerous to have this right now because Once they're locked in, like you're basically renting from them to run your company, right? And you don't want to do that. That's it for today, and we will see you tomorrow.