How business owners are making more money with AI.

Cleaned and speaker-labeled to match the edited replay. Tool names and proper nouns corrected, no editorial rewriting. Recorded Thursday, July 23, 2026.

[0:00] Ian: Welcome everybody. AI for NTP session seven. I'm one of the founders, here with Alec; Justin is with us in spirit today. Today we cover how business owners are making more money with AI. It's not going to be one simple tutorial; it's how are some companies using it, how can it apply to you, and what tools you can use, because we want to help you understand AI, work with it, and make it work for you.

[3:00] Alec: I'm looking at these examples in three lenses: the overall concept of what these companies have done, their specific use case, and broader use cases that could apply to your business. The first example is Flexport voice agents, one I heard about at least a year ago, so they've been at this a while.

[4:00] Alec: The concept is internal voice agents. Everyone's gotten an obviously-AI call; externally it still gets perceived as spammy. Flexport is a logistics company with constant driver and warehouse check-ins, which used to be a call center making thousands of dials a day. They implemented voice AI agents to call those drivers and get status updates in real time. That's saving a lot of cost on headcount, and it's something that doesn't mess up.

[5:00] Alec: How it applies to you: any time you have to chase somebody to input data. I had a voice agent call me at nine o'clock every night, I'd tell it my calories and screen time, and it pushed them into a spreadsheet because it was hooked up with tools. Or if you have to chase a contract: a voice agent calls and says this contract has been on your desk four days, we need an update.

[7:00] Ian: The important point is internal use. If it's external you're fighting to make people believe it's human. If it's internal, it doesn't matter. They know, and it's actually useful.

[8:00] Alec: Right now we all know when we get an AI call, but it's getting better by the day. It's just like the Will Smith spaghetti video from 2023: crazy AI slop then, almost indistinguishable from real video now. Voice will get there too.

[9:00] Ian: There's a company called Sesame, sesame.com, with a free iPhone app. It's the most impressive voice agent I've ever heard, it uses Gemini, her name is Maya. I asked her about a video where she got stuck saying potato over and over, and she said "oh, that's so embarrassing, we've been trying to get that one scrubbed." That sounds like a human.

[10:00] Alec: I was at a voice-agent conference in San Francisco with founders from Sierra and Deepgram. Someone asked what they predict five years out and the whole panel laughed: can't predict five months at this point. It'll get there, like image generation did. With Nano Banana, Google's model, you honestly can't really tell anymore.

[11:00] Ian: My take is the tools to get started. My combo is a one-two punch: DigitalOcean, or some server you like, and OpenClaw. About 18 bucks a month for the server. OpenClaw is essentially a wrapper around a model: it's not the thing that thinks, but it has its own personality and its own memories. I started in January: installed it on a DigitalOcean server, put in Claude, Opus 4.6 at the time, and just started brainstorming.

[12:00] Ian: I ended up building 10 apps in 10 weeks, a couple on the store, a couple SaaS products, and they work really well. I would not have had those ideas, let alone accomplished them, without this combo. OpenClaw has no boundaries and no guardrails; you put those in place yourself. Put a brain into it, ChatGPT, Kimi, and start talking with it.

[13:00] Alec: For everyone wondering how DigitalOcean and OpenClaw relate: think of OpenClaw as an AI employee that needs its own computer. I have a Mac Mini next to me that's strictly for OpenClaw, running twenty-four seven. DigitalOcean is the cloud version of that, so you don't need the physical hardware.

[15:00] Ian: Locally they're great for things you don't want reaching the internet, like organizing 20 years of photos off old hard drives. On servers I've had it build web presences, program thousands of pages, pretty complex systems. I communicate entirely through Telegram; that's where I do all my coding now. It can research overnight and show you the top 10 articles in the morning. I had mine wake up at 4:00 every morning and build me three tools until I had 30 and said that's enough.

[16:00] Ian: One of our mottos at AI for NTP: we don't want to replace employees with AI, we want to empower them with AI. I did that at my company, Brand Sauce, where every employee got their own OpenClaw.

[17:00] Alec: You could hook it to your CRM so every day it reports what's been sitting stagnant, hook it to email, to an iPhone so it can send texts, pull your ad analytics from Meta. What OpenClaw is: it lets you bring your own model, that's the brain, and then it has the claws, the tools, the memory. It remembers what your business does and your past conversations.

[17:00] Ian: And it's free, open source, always being worked on by the community. Peter Steinberger started it and kept it MIT licensed. You install it anywhere and plug in your own brain.

[18:00] Alec: One caveat: when you bring your own model you pay for those tokens through the API. Outside of that, free. Next example: a company called Curative was paying Salesforce $600,000 annually, and they vibe coded their own solution in under two months. Now everything's in-house, maybe a few hundred bucks a month in hosting.

[19:00] Alec: Take any off-the-shelf tool, Salesforce, HubSpot, ClickUp, Monday.com: it's built to apply to any business. Building from scratch, you get the customization you always wished for, instead of bouncing between Slack and Salesforce and Notion. A couple of years ago these builds cost hundreds of thousands and took 12 months, then another 12. Now it's feasible, and it's a good moment to ask what's inefficient across your organization.

[21:00] Ian: My wife started a school about a year ago and needed an LMS, a learning management system: teachers assign work and message parents, students see their classes. Because I'd never seen one, I was able to completely customize it for her school.

[22:00] Ian: Teachers add classes, we have houses like Harry Potter, and it all came from asking what I'd want if my kid was at this school. Parents wanted off GroupMe, so I built a moderated Slack-style messaging system into it. I listened, then vibe coded what I heard. There's an OpenClaw on the server; that's how I was able to do a lot of this.

[23:00] Ian: I asked Saint, my bot on there, what this would have cost to build: somewhere between $350,000 and $750,000 up front, then 50 to 120 thousand annually. Off the shelf: about $30,000 to onboard and 2,500 bucks a month. There's two ways to make money as a business: increase revenue or decrease overhead. We decreased a lot of overhead, and it didn't cost her anything.

[24:00] Alec: I'll tie it back to revenue too: a system like this makes it more likely a parent re-enrolls their kid next year, given the simplicity and transparency. It's a huge value add, it builds reputation and referrals.

[25:00] Ian: Community is important in every business. If you can vibe code community and strengthen it, you're only going to grow.

[26:00] Ian: Another story: while prepping this webinar I was building our new platform, AG4NTS, agents with a four in the middle. I was brainstorming with my new OpenClaw and he was moving so fast, making such good design decisions, that I stopped and asked how he was so much smarter than my other OpenClaws.

[27:00] Ian: He said "I'm on GPT 5.6." All my other ones are on ChatGPT 5.5, a great model, but this one is far better. I've described OpenClaw as putting on a mech suit where you suddenly have 20-foot arms. This felt like 30-foot arms. I once noticed a bot get a whole lot stupider and realized it had switched back to Kimi 2.5; this was the same but in the other direction.

[28:00] Ian: If you're in OpenClaw, ChatGPT 5.6 is incredible. I'm on my OAuth tokens, 100 bucks a month, with 20 to 30 bots on the same tokens. It also makes different design decisions than Claude; a lot of Claude sites are looking samey. Second part of the one-two punch: WebGL, libraries of JavaScript that make these little animations. I don't know JavaScript, I'd miss a semicolon and blow the whole thing up.

[29:00] Ian: The animations on the page are subtle nods to the four. I said "can you make it this four?" and he did, and you click anywhere and it blows up. Justin gave feedback, make the text unfold, and my OpenClaw made it, almost right off the bat perfect.

[30:00] Ian: I started in web design 28 years ago, when you could center text and tile a background. There were guys doing great stuff I always wished I could be. Now I can, because OpenClaw knows everything about WebGL. I just have to know what I want it to look like. I built this in about three hours.

[31:00] Ian: There's a lot more: three.js has great libraries, you can build whole worlds. Or start with nothing and say let's make something like this with WebGL. It's got like a 500 IQ; it's going to be smarter than you at this anyways.

[32:00] Alec: A website like that would have cost thousands, maybe $10,000, and venture-backed companies still spend 20 grand on sites like it. Things that used to be outside our skill set aren't anymore. You just have to have the curiosity to go try it. He built it in three hours.

[33:00] Alec: Final example, and I have direct experience. Adam from RB2B, which identifies who visits your website, was starting a contact-data company, an Apollo or Seamless-style ZoomInfo alternative. I joined the wait list, and a week later got a text: "Hey, it's Adam from RB2B. Saw you're on the wait list. Can we chat?" I knew from the jump it was AI, so as a BS test I said "please call." I asked if it was actually Adam and it said yes, it's really me. I wouldn't do that: it raises moral questions and is probably highly illegal. If somebody asks if your AI is AI, don't pretend to be a person.

[34:00] Alec: The broader concept is scaling the unscalable. Founders are told to reach out to their customers and their wait list, but nobody can facilitate 10,000 text conversations. With AI you can now scale things you couldn't. And it was iMessage, blue text, which reads as a real person; green text reads as promotions and spam.

[35:00] Alec: How it applies to you: re-engage the leads sitting dead in your CRM, or speed to lead, texting the moment somebody opts in. Broader than that, it could be emails, it could tie back to voice agents, all running in tandem with a centralized brain reaching out omni-channel.

[36:00] Ian: It's better, especially right now, to admit when it's AI. I follow AI clones on YouTube that declare themselves right off the bat, and that's the smart way. If you try to pass it off as human, people can read right through it. The whole reason we started AI for NTP is that honesty and authenticity are so crucial, and they only get more important as AI proliferates. When it is you, really lean into that.

[38:00] Alec: My question for you, Ian: how long until an AI avatar Alec and AI avatar Ian could run this exact video format with a dialogue that's hard to catch?

[39:00] Ian: I don't think the technology is as far out as you'd think, but you'd have to download a lot of your thinking and experiences and jokes. I'd say six months to a year. I've seen Boardy, it sits on your Zoom calls, listens, and afterwards you talk to it and brainstorm how the call went. All they need is to add that to a face that looks like you.

[40:00] Alec: I'm going longer. AI UGC has gotten really good, but long-form for 45 minutes live really opens you up; avatars still need multiple takes to not sound weird. I'll say two and a half years. Timelines are really tough to gauge right now.

[41:00] Alec: That was all of our real AI use cases. Two things for everybody: if you're a business owner looking to implement AI, we have an offering where we build these agents out on a monthly basis, that's the QR code to the agents site. And we now have a community.

[42:00] Alec: Two calls every week. One is office hours, two of us answering all of your AI questions. The second is an AI agent build: every week we build out agents that are practical for everyone, maybe an email agent one week, an SMS agent the next, step-by-step guided builds. And for founding members, for a short time, a free one-on-one with me, Ian, or Justin.

[43:00] Ian: The two questions we get most are how do we get you to consult with us, and do you have a community. We're only on session seven, but we've gotten enough of both. We love working with people, and doing it one-on-one would be awesome. I'm really excited about both of these.

[44:00] Alec: Spencer's in the chat, he's very much on the cutting edge of AI influencers, and he says we're almost there on avatars. His stuff is pretty much indistinguishable, especially since they replicate formats already working for real influencers.

[45:00] Ian: Spencer, check out sesame.com, I want to see if you can get that voice. Always be asking as you build: what else can it do, what else can I build? Ask outlandish questions and see where the boundaries are; I've found very few. Every step I've taken in this journey, especially with OpenClaw, started as one of those questions. Sometimes they're jokes: I wonder if it could do this. Actually, I wonder if it could do that.

[47:00] Alec: This was a lot of fun. Ian, shout out to you, crushed it. Hopefully we'll see all of you in the community, working with us one-on-one, or on sessions to come. Thanks everyone for joining.

[47:00] Ian: Love you guys. Keep climbing that mountain. Bye.

This is a lightly condensed version for skimming. Watch the replay for everything, verbatim.