Hank: Okay, two questions for today. Is being secretive better for marketing than being loud? And what are the updates you need to make to your product and website to capitalize on agents?

Gonto: I have a third one. Is my bed being made up better than not being made up? I no joke almost said question number three.

Hank: When's Gonto going to make his bed? Wait, are you claiming that that bed is made right now?

Gonto: It is, yeah.

Hank: If you're not watching the video... Look, wait, look, look. Gonto's bed is not made.

Gonto: It's fine, look, you can move it and open it.

Hank: The covers are on the bed, correct. You just unmade your bed.

Gonto: I don't care.

Hank: In your demonstration. Yeah, it's fine if you don't care, but you can't claim that the bed is made now after you've unmade it.

Gonto: This is the closest you're going to have. But let's start with the first topic.

Hank: This was our attempt, by the way, we were like, hey, let's have like a good hook and see if we can improve the retention on the first couple of minutes. We've chased them all away. That's OK. The new viewers are gone. Only the real fans have stayed. Let's jump into this Ox Alpha. Let's go.

Gonto: So if you've been on Twitter these past five days, you've probably read about Ox Alpha, where it's a new model that is being served basically in all of the open model companies like OpenRouter has it, OpenCode has it, Nous Research, the Hermes one has it, and everybody basically has it. And I've seen a lot of people trying to guess on Twitter what it is. Some people talking about how good it is, how people talking about how it sucks. Some people explaining about how it works differently than others. Some people are trying to guess that it's the new Gemini. Others are guessing it's GLM. Everybody's guessing something different. But to me, what's fascinating is that they decided to go secretive for five days, whoever this Frontier Lab is, and everybody accepted it. And it's out there in that secretive way. What's your take? Do you like it? Not like it? What would you do, Hank?

Hank: So I want to understand because I guess I wasn't on Twitter the last five days. What you're telling me is some lab, we assume it was one of the Frontier Labs.

Gonto: Yes.

Hank: And put out a model. Nobody's claimed it. It's just out there. And so people are like, you know, OpenCode, OpenRouter. They're like, hey, you can use it now. Try it out. It's so interesting. It makes me immediately think that it's going to be a Meta or a Microsoft or somebody who doesn't have a reputation right now for having great models. They want whatever you know about them to just be ignored. Like, hey, don't worry about who this is from. Just evaluate the model for its merit. And if that's the case, that's going to be really good because then everyone's going to be like, this thing's amazing. Now, on the flip side, if this sucks, if everyone's like, this model's terrible, they don't even have to claim it. They just like go, OK, interesting. And then they just, you know, take their notes and they come out with a new model in a month or something. I don't know.

Gonto: I love it because everybody on Twitter has been talking about it. And what I like about it is a couple of things. Number one, there's multiple people trying to guess. And that people are trying to guess are just putting names or at least they are talking about the model. So even if they fuck up the guess, like if it's not the right company, people will know that somebody is talking about Ox Alpha and what it is. So I think everybody guessing, I think, is good for whoever builds the model. Number two, the model is performing, it seems, pretty well. And what I like about it is there's a lot of influencers who are full of shit. And what I like a lot about this is that people who are trying to guess based on evals and stuff like that, they're starting to show up on Twitter. So it looks like it's not just influencers, like your typical influencers talking about this, but actually very, very smart people who are just trying to guess because they think it's a fun game. And you're getting those folks to try your model.

Hank: You showed me there's a big Polymarket on this.

Gonto: Exactly, there's even Polymarket, like that's how big it got. Like people are trying to guess. Two of my friends were like, I'm sure it's GLM because it's been trained there. Some people are now saying, for example, that it's not from GLM, but it's trained based off of that. I like the theory that it has to be open weights because otherwise the Trump administration will have to approve it. So, so many theories, so many people talking about it. I just thought it was the best AI marketing move of the year. Like I see companies doing launches, some launches get 1000 views, some get 2 billion views. But in this case, because the model is good, they just got people to try to guess what it is. They got people to talk about it. And everybody's trying it out because it's free. So I honestly think it's like, to me, it was the best marketing trick of the year.

Hank: Yeah, it looks like I'm looking at the Polymarket now. It looks like the money's on that it's Z.ai, which they came out with like a pretty hot model a couple months ago.

Gonto: Z.ai is a GLM company, but a lot of people are saying that it's not really from them, but rather that it's somebody that's trained a new model based off of their open model, which could be true. But I've never seen anybody do something like this. The only closest thing to this was Cursor, that was basically publishing their own new models secretive inside their app, but not somebody actually giving it for free to all of these OpenRouter or routers in general so that people could try out. So whoever you are, I'm blown away. I think you've done a really good job.

Hank: Very interesting. It could also be, you know, is it a, I don't know, is this some cyber attack that's about to be unleashed on everybody? Who knows?

Gonto: People are using 1 trillion free tokens per day of it.

Hank: Wow, crazy. Yeah, I agree with you, though. Like the stealth move is interesting. It's not going to be one of the big players. It's going to be somebody who people perceive as behind or it's a new player or something.

Gonto: Let's move to the second topic. Brought up a tweet from Zeno. Want to talk a bit about it and what's your take of it?

Hank: Yes. OK, so Zeno from Resend brought up about how something has changed, how he buys software and he's reflecting on this and he's reflecting that at work, I only buy services with an official MCP. And now at home, same thing. He likes Strava because he can ask Claude about his runs, whatever. And he says, I used to care about features. Now I care about interacting with my data. That's a really interesting shift. Now he's at the forefront, right? This guy's super AI-pilled. His company's riding the AI wave. So not everybody's going to have that same mentality. But if he's at the forefront, it stands to reason that eventually people are going to catch up to that type of thinking. And Resend has also done very good at making their site very easy for agents to read and use. They're not an AI company, but they're reaping all the benefits because they've made their product the best email to be used by agents. And they've employed a lot of great tactics, which we can talk about. But I mean, first that philosophy, like how long do you think it takes for the broader market to become like Zeno, to care more about interacting with their data than they care about features?

Gonto: When we talk about the broader market, I would separate it. And for us, it's like what we care about, number one is developer tools, I think. For developer tools, I think it is happening right now. So it's like right now, everybody's becoming this similar to Zeno on the frontier. So I think if you're a dev tool, you need to think right now. And we'll talk about some things you can think about on agent experience and agent discoverability and what things people care about for their agents. I think for B2B in general, it's going to happen in the next two to three years, probably, as more and more people are going to start using, I don't know, GrubBot, for example, launched last week. GrubBot, to me, is like a Hermes or OpenClaw for non-technical people. And they just started deploying it for companies. And I think if GrubBot starts working in companies, companies will need GrubBot to be able to integrate to existing services and tools or something similar to that. So I think in the next two years, that's going to happen. And most companies are going to anchor it. For B2C, I think it's going to take a lot longer, maybe five to 10 years, just because it's going to be really hard to get everybody to think about it. But for us, I think that we focus mostly on developer tools. You have to be thinking about the things that people care about now. Like I think for DevTools, UI is dying.

Hank: Yeah, for DevTools, a lot of the users don't have eyes. The necessity of having data interactable, which isn't a word, but you know what I mean? Having data available is the new important thing. I don't know. I think we're starting to see a slew of new tactics for making things easier on agents and people's agents, which I'd love to like, I think Clerk, which you advise has done a lot of good stuff with this too. And we've talked about some of it before, but I'd love to rehash, you know, what our audience can actually do today to capitalize on this more. But what were you going to say?

Gonto: I agree with that. Like, at least to me, if you're a DevTool, things that you should be thinking about now is number one, have you created skills on how to implement your product for each of the different frameworks, languages, et cetera? If you haven't, I think that's step number one. There has to be skills that explain how to connect and integrate to your service. Number two is MCP. Like MCP is back. So I would say you need to find a way to have an MCP that wraps your API so that people can use it. Some people hate MCPs.

Hank: Explain the hate on MCP. Like I've seen an oscillation in the last couple months, but I'm like, it seems like people still really want the MCPs. So I don't get the MCP hate.

Gonto: The problem with MCPs was that an MCP, for example, has a lot of tools, a lot of things that can be done using it. And before, all of the MCPs and tools were loaded when you were starting a chat. That meant that they used a lot of your context. So a lot of their context was already basically dirty because of that. And that meant that you could do less things or the model could remember less things based off of that. Now MCPs, thanks to changes in Claude and Codex, and Hermes does the same, are loaded on demand. So basically, when you want something, you describe what you want and it will load on demand to the context that MCP. So with that, it's not that bad anymore because it doesn't clutter your context. And therefore, I think that's one reason that it's back and it's working. But I think doing a CLI also is something that makes a lot of sense just because LLMs are non-deterministic, which means if you write a fantastic skill, it might happen that your LLM doesn't fully follow the skill. It will try to, for sure, but some cases it will follow and some cases it will just fuck it up. So I think having a CLI basically enforces the use of the skill or something similar. So what you could do if you're a dev tool on the CLI is something like, okay, use this skill and then check if it has used it. And if it hasn't, do something else or try to load another skill or do something else about it. But I think having a CLI basically improves the likelihood that things work from the get-go on the first run versus, oh, there's a problem or something like that. But I think that's why I would recommend doing a skill, doing some MCPs on top of your APIs, doing the skills on how to use all of this and how to implement it. And then also have something around instead of sign up to your platform or instead of in the onboarding where you need to click things, having something where you can just copy a prompt, like copy a prompt, which maybe installs a skill or installs a CLI. But copying a prompt on, ideally, you can sign up to your dev tool through the API. If you can't, then maybe instead of doing that, at least once the user is signed up, you can change everything from the dashboard or anything from the API just using that MCP or doing calls. So I think copying a prompt is also another thing that to me is very important to have inside of your product. And then, of course, it's talking about all of these things. Like before, people on the homepage talked about the features, showed pictures of the UI and stuff like that. I think if you're a dev tool, your UI matters less. And now what matters is, do you have a CLI? Do you have the MCP? Do you have these skills? Are you talking about them? Is it easy to interact with others? And I think from a marketing perspective, we haven't thought enough about all of the things that we need to highlight now on homepages and on websites, which we didn't have to before.

Hank: Yeah, I don't think I have anything to add. That was great, very informative. Thank you.

Gonto: Tell me about... You also shared a reel about the direct mail and you told me you're doing direct mail. Tell me a bit about that. Like what was that reel and what are you doing for direct mail?

Hank: The reel was this guy basically saying, AI companies have too much money because he got this letter. We'll show it on screen that like you unfold it and it turns into a little cube and then it starts talking at you and it's by ElevenLabs and it uses Sir Michael Caine's voice and it's kind of silly and possibly annoying, but also pretty interesting. I think if it hits the right people and yeah, this guy's complaining about it, sure. But I bet that hits with some people. Some people got that and they were delighted and they hear Michael Caine's voice. They were like, wow, that's amazing. It's uncanny how well it sounds like him and it probably gave them ideas on how they could use voice agents and it feels like good marketing to me. I don't know. Did you watch it? What did you think of it?

Gonto: I did watch it and I liked it a lot.

Hank: Even though this guy...

Gonto: It's going to work, right?

Hank: I think it's going to work.

Gonto: Exactly. It's like you're going to listen to the audio. You're going to like it. As you said, you're going to think out ideas, but overall it's something that surprises you, makes you think differently and at least like if it doesn't work, meaning like you don't implement it, maybe because it's so creative, you share a reel about it and then other people will watch it and then talk about it, which is still pretty awesome, even if those people are not going to use it.

Hank: Yeah, I mean, this guy got, I mean, it's easy like engagement bait to hate on AI on Instagram, but he got 5k likes for this, you know, hundreds of thousands of views and yeah, I don't know. Now you mentioned that I'm doing a little direct mail too. I'm making a postcard. By the time this gets published, everyone should have gotten their postcard who came to the Code to Market Summit last year and I just thought that would be a fun way to like invite them back. It's got a QR code to a private hidden link on the site where you can actually register because I try to keep the Code to Market Summit invite only. So it was fun and I made the postcard. I made the design with AI, you know, my EA's sending it out. The hardest part was collecting people's addresses.

Gonto: I'm a big fan of direct mail now, mostly because everybody's back in office. So I think you can just send it to the office and people will see it. And with outbound doing much worse, sending it to an office, direct it to a person so they see it. I think it can call people's attention and work. The other one that I've seen in the past that I really liked was what Browserbase did where they have the team actually walk to other companies in SF with custom Legos or Legos, I don't know how you say it. And they would literally, Legos, so they would literally just knock on your door, on your office and then give it to you. So they force you to talk to you and of course they can do it because they sell mostly to like SF companies. So they can do it. Maybe in other cases you can't. But this idea of it's not just direct mail, it's like we're knocking on your door. I loved it. I thought it was absolutely incredible.

Hank: I've seen this. People have done the, hey, I'll bring doughnuts to your office if you, you know, just tell me where your office is and let me bring your office doughnuts. And that's it. Like you don't have to go and make a pitch. They just, it's like, all right, I'm just going to pick up some doughnuts and drive over there.

Gonto: Exactly.

Hank: And it is so much more effective. It's so much harder to stand out in someone's inbox. I mean, the AI SDRs are ruining everything. You can't stand out anymore.

Gonto: I think outbound is dead or in its progress to dying. So yeah, I think direct mail is a new way to stand out. I agree. With that, I think we can call it a day. Thank you for joining and listening to us and hope that you got hooked by our new hook questions. You let us know.

Hank: Thanks everybody. That's the pod.