What it takes to build a go-to-market AI stack on Claude yourself

The short answer. Claude with a few MCP servers connected is a working first level, but it is slow and gives a team no way to measure the output. Finishing the build takes four more levels: fast search across every conversation the team has, a work surface reps adopt, an agent for each person that runs skills and takes their feedback, and agents that act when an email lands or a meeting ends. Day AI provides the context graph and agent control plane to quickly and securely upgrade your company brain, so a team spends its effort on skills and workflows built on top.

What the video covers

• 0:31 In the deals Day AI sees, the most common competitor is a team deciding to build the stack itself.
• 1:09 Level 1, Claude plus MCPs: it works, but it's slow, there's no measurement, and weak drafts end up in Salesforce and in front of prospects.
• 1:39 Level 2, one searchable record of the business: teams stand up Elasticsearch, Turbopuffer or Pinecone, get Gong and Slack in, and stall there; until everything is in, the MCPs alone serve better.
• 3:54 Level 3, the workflow layer: reps only adopt a consumer-grade product, so whoever builds it takes on every bug, feature request and point of UX confusion as a full-time job.
• 5:03 Level 4, agents for each person: skills that run on someone's behalf, and a way for that person to say "do it this way" and have the automation change. Asking Claude to fill out a form and waiting 30 seconds is no gain over typing into Salesforce.
• 7:18 Level 5, event triggers: run something every time an email arrives, a meeting ends or a customer posts in Slack. Sandboxes and cron jobs can get there, at a heavy cost to build and run.

Excerpts

[0:00] Depending on the end state that you are after, a very much scarier thing to take on than folks realize. Maybe we should just talk about it. Go in eyes wide open, fully understand what you're actually getting yourself into, and what this actually can and should look like. I have been in Day AI looking at our coolest companies, customers, prospects, et cetera, and we have one competitor, and it is not a vendor. Ooh, curveball. Curveball is right. I want everybody to believe in themselves. […]

[1:10] Let's get into it. The current effort for all of these customers right now is something around Claude as the work surface, MCPs connecting in. Here's the first question: can they do that? They can definitely do it. It's totally doable. It'll work. It'll be slow. Yep. And they won't have any way to measure it and show for that. It's going to be super slow, and people are going to be producing slop that then goes into Salesforce and goes out to prospects. Doesn't really get you anywhere, right?

[1:45] Okay. So, level two: Perplexity of go-to-market. This is the Jack Dorsey single-brain idea. Look at all the information you're generating just by doing your work. […] What are we seeing? We're seeing folks get not very far. We are seeing folks spin up something, Elasticsearch or Turbopuffer or Pinecone or something like that. They can get Gong in. They can get Slack in. That's kind of where it ends. Exactly. If you're going to do that, you're better off just probably using the MCPs.

[2:27] If you can get all of the key information in by default, and it's all hooked up the right way for the agent to run around it, and it's all structured as natural language, then you have the Perplexity of go-to-market. It just screams. It's super fast. You think there's something better than the Perplexity of go-to-market? Yeah, I guess it would be the Cursor of go-to-market. Nailed it.

[2:51] Level three. Now, this is a form-fitting user experience that's modern and just absolutely simplifies your workflow. Allows you to take actions right in place. Just beautiful, fast, consumer-grade, with the AI as the principal element within all of it. The go-to-market version of that would be what? You would want an intuitive starting point that may look like CRM but probably looks like Claude, or at least you interact with it like Claude.

[3:25] And I guess if we took the Cursor image literally, you would be able to look at a view of accounts, organizations, deals, whatever, and you could pull up your agent and ask it to rearrange it or cut you a view or something like that, and it would do it really quickly. It would be beautiful. You would be able to come back to that. The overall frame of the whole product experience would need to be this consumer-grade, hyper-modern thing, otherwise nobody's going to use it, right?

[3:57] If you want a workflow layer that your broader sales team is going to actually adopt and get some value out of, and actually have increased productivity, that needs to be a consumer-grade product. You are signing yourself up to take every bug, every feature request, every UX confusion, and get it absolutely perfect for them. Otherwise, they're not going to touch it. Oh, and news flash, that's somebody's full-time job. That is somebody's full-time job. Which, I guess, taking user experience feedback from sales reps and spending your entire waking life focused on it, that's some people's idea of fun, but not everybody's. No.

[4:35] […] So now they have the perfect, super-fast search engine, Perplexity, context graph piece, and they have the native creating emails inline, super smooth, updating CRM data, like everything is super Cursor of GTM. Let's grant that they've gotten that far. Okay. […]

[5:08] All right. So let's tackle it. Level four. Mhm. OpenClaw of go-to-market. You would have to build some sort of system executing skills, let's say, of agents, let's say, on behalf of the people who are in charge of that work. Mhm. And it's by person, and they have taste and opinions, and they're going to need to be able to weigh in on that automation and say, "I don't like it like that. Do it this way." Yeah. The concept around OpenClaw and the way that it handles this was very elegant, and happened to be the way that the AI works. But that's what you would have to build.

[5:45] If you're scared, you're paying attention. Yeah, exactly. It's a scary thing. It should scare you. I'm in danger. But you have to remove the human from the workflow somehow, or else we're in the same boat we've always been in, which is a workday. A workday, or a really long workday, but you're still limited by human capacity at some point. And you fall into that trap of everybody's using the AI stuff, they're working more than they were before, and we don't really see any obvious difference or benefit to that. If there's an obvious difference, it may not be an improvement.

[6:25] Yeah, that's not really the goal here, to have AI just having us do more work. The goal is to have the AI allowing us to do more output with less. Telling Claude to fill a form out and waiting 30 seconds is not an improvement over filling out fields in Salesforce. Yeah. Yeah. It's not any faster. It's not any better. It's worse. Well, you get to spend the tokens. I mean, you get to spend the tokens, but what are we doing if we're not making progress? […]

[7:12] Okay, so maybe you can spin up a sandbox environment. You could definitely do some sort of cron thing. You could build that. You get into the event triggers. Now it's kind of a pain in the butt. Now you're back to Gmail: every time I get an email, do this. Every time a meeting ends, do that. Every time a Slack with a customer comes in, do it this way. Theoretically doable. Nightmarish. I mean, we're doing that.

[…]

[8:15] So, you've opened up this whole world of, like, there's all these things that I actually really would love for somebody to do for me. You would never hire somebody to just scan the internet for podcast opportunities for you, but you can make an agent to do that. It makes total sense to do that. Now you can have a whole team doing those things. So it changes the way you think about what can even be done.

[8:45] But that's where we are. We're, like, level five. You can just grab it. It has all of that. It has the Perplexity of go-to-market, the Cursor of go-to-market, the OpenClaw of go-to-market, and then this agent virtual worker thing […] Like, it just works. It's crazy. I couldn't do my job without it. […] It's not even hyperbole. It's fact. I would not want to do any job without it at this point.

[9:22] I mean, even this stuff we're talking about, we're talking about this because this is the stuff that prospects are talking about and customers are talking about. Therefore, it's what you and I are talking about. Therefore, it's what we're on camera bringing to the world. It's pretty cool. 100%.