Sell Your API to AI Agents & Make SERIOUS Money in 2026 (full guide / beginner friendly)

0:00What's going on, guys? This is Oliver, formerly from Response AI and a few other software tools that I've since exited and now running roosevelt.dev, which is where I teach people how to code and make money with software, and running paperschedule.com, which is my new SaaS. In this video, I wanted to talk a bit more again about APIs, but specifically how to sell to AI agents, right? So, I don't mean This is an interesting one. I

0:24don't mean selling agents, right? Because agents We all know what an AI agent is. We all know that you can make money with them. This video is about how to sell to AI agents to actually make money, right? Um we're going to be talking about APIs. We're going to be giving you um actual ideas on how to make money and how to sort of take advantage of this, but it is going to be a giant market. Now, for the entire history of the internet, your customer is a human, right? They just buy your software. But, a new customer is showing up, and it is an agent. It's not a

0:53person. It has a wallet, billions of customers, millions of wallets. Almost nobody is actually building to sell to AI agents just yet, right? So, your customer is a machine, and it wants nothing you built your website for, right? So, people now have AI assistants in their pocket that books dinners, uh compares vendors, file support tickets, buy software, finds leads, whatever it may be. Businesses are spinning up agents that handle procurement, research, sales, whatever

1:24it may be. These agents are just They're coming out of nowhere, and they are constantly constantly getting uh sort of more reliable. They are making more money for people. They are doing more stuff, right? The people thinking clearly about this believe that agent traffic will outnumber human traffic before long. So, let's just really, really like go to the basics, right? If I say um to Claude or, you know, to one of my AI tools, "I want to build a software that allows people to manage their customer support for

1:55their bakeries, right? The problem is that obviously you can build a software that does that, but we need an we need an agent to do that, right? We need um uh an email inbox that is the agent, right? And that's Agent Mail, for example. Um we need um a tool that actually manages what the replies are. So, if someone complains, how do we know that that's a complaint? All these different

2:19things. Selling to agents does not reward the prettiest landing page or the biggest ad budget. It rewards whoever exposes a useful capability, like a useful agent or tool to other agents, right? So, I know this is a bit it's a bit hard to explain, but obviously um just bear with me, right? What an agent actually wants is a human customer um or comparatively, [clears throat] a human customer wants persuasion, a demo, social proof. An agent just wants three things, right? Clear capability of what you've built, permission to use it, and then enough trust signals to know it will not get burned. So, it doesn't read your read your story, it doesn't look at

3:01your face, it doesn't read um your blogs, it doesn't watch your videos, it just reads your structure and calls your endpoint. So, if an agent can't understand what you do and act on it through code, you are invisible, right? Now, if you go to any software, so if you go to Zurnio or you go to Postees, you go to Lead Magic, all of these tools, their landing pages literally say um that they have an MCP or that they have a way for agents to talk to them, right? So, a lot of these big companies are already setting up for um you to basically never have to worry about

3:36buying it yourself, right? In other words, an agent will buy it for them. So, every step how an agent actually buys. So, it finds reads documents and uh pricing, right? It doesn't browse, it doesn't have decisions, it doesn't have opinions. It evaluates, right? It passes capabilities, limits, policies, whatever it may be. It checks if they can trust you. Do we have proof of identity? Do we

3:59have compliance? Are we safe to use? Then it transacts, right? It pays, it books, it subscribes to the software, whatever it may be. And then it recommends. So it tells other agents what worked. And that word of machine loop will decide who wins. Now, the rails already exist. You don't have to build them. The reason this is a solo founder opportunity is it's not a raise $50 million

4:24opportunity. Um it's basically that your job is just to assemble it the same way you would wrap any other API that we've talked about. So there's three rails. You've got payments, identity, tools. None of them are yours to invent. So let's talk about rail one. Agents can already pay you. So this was solved pretty quietly last year. This is what I mean. It's not like a very common thing that's being talked about. OpenAI and Stripe built the agentic commerce protocol, right? It powers instant checkout inside ChatGPT, right? Etsy has it live. A million Shopify merchants are

4:58coming on. OpenAI takes about 4%. Google and Coinbase built the agent payments protocol. So for agents paying on a user's behalf with strict spending mandates. The one to watch as a solo founder is Coinbase's X402. So any API endpoint can charge a tiny fee in stablecoins. No login, no card, whatever it may be. Now the result is an agent can hand you money in fractions of

5:23a cent millions of times a day. Rail two is that agents already have inboxes. So agent mail gives each agent its own inbox. So like we talked about the bakery example, if you were talking to customers, you'd need an agentic email inbox. You can't really wrap, it's hard to wrap like Gmail into an agent. It's hard to put agents inside Gmail, whatever it may be. There's no human clicking buttons. There's no actual inbox itself unless you set it up. If Agent mail is just it sends, receives, creates threads, and reads emails on its own. And it can even use that inbox to sign up for other services and pull

5:58verification codes, etc. Right? So, they raised a lot of money and the demand is real and there's a free tier for you guys to try out. Let's say, for example, I'm building Paper Schedule, which is obviously, like a sales agent. I need Agent Mail to set up inboxes. Um but when I asked Claude to um find me a software that gave me inboxes, it it told me Agent Mail. So, you see how I

6:26didn't do any research on Agent Mail. It's just that my AI, so Cursor or Claude or ChatGPT, it told me to use Agent Mail. So, you see how Agent Mail is selling to other agents. So, that's what I mean by selling to other agents. Agents can reach you through tools. So, this is a bit more complex, but MCP is the new front door. So, instead of an agent trying to click around your website like a confused human, you

6:50expose an M- MCP server. So, excuse me. An MCP server, you can think of it as like a clean menu of actions that your software offers. So, you can search customers, create an invoice, issue a refund, update a ticket, whatever it may be. The agent reads the menu, picks the action, and calls it. And that menu is becoming the real front door of your business and you can build

7:14it before competitors do. Um Zeno have done this, uh Posties have done this, Lead Magic have done this, all the tools that I talk about on the channel. And there's two plays, right? You can pick one. Defensive or offensive. Play one is that you make what you already sell viable by an agent. So, it works if you have an existing business, you capture agent traffic while your competitors are still wondering while why their human conversion rate is sliding. Play two is that you build the

7:44picks and shovels for the agent economy. So, where a fresh API first solo founder moves fast, which is what we talk about on this channel, the whole economy is missing supporting tools. So, every gap is a product. So, play one is that you can make yourself viable. So, three moves this week, expose an MCP service so you let agents do business with you through actions

8:07instead of guessing at your buttons. You add discovery file. So, all software that it that that are doing well right now, they have an LLMs.text, right? So, Granola, they all have these um they have these tools, they have these sort of like documents that tell the agents how to use their software, right? And it's agent readable, pricing, policy, etc. Then you can plug yourself into a checkout protocol. So, ACP, X402, or AP2, depending on your model. An agent can now pay you without a human's

8:38opinion on it or a human in the loop. Play two idea one is an agent ready um software in a box. So, this is for small businesses who have no idea what is coming. So, almost no local businesses have any idea that agents are about to become shoppers. So, you be the person who makes them ready. You take their website, you generate the discovery file, the agent readable pricing and policy page, and a simple MCP server exposing their core actions. You can do this to other software companies, you can do this to commerce companies,

9:07whatever it may be. The math is that cost per per customer in pennies, charge a setup fee, then $39.99 per month to keep it current as agents and standards change, right? The pitch is, do not become invisible when agents start buying stuff. Play two idea two is a paid tool agents call directly. So, the purest version of selling to agents. So, this is like Zeno or like Postie's or like Lead Magic or like Granola, whatever it may be. The agent is the customer, it pays you per use. So, pick one useful, narrow capability, like a verification check, a niche data lookup, a formatting job, anything an agent needs mid-task and

9:48would happily pay a fraction of a cent for rather than figuring out itself. You wrap whatever does the hard part underneath, add your own logic on top, and then charge per call using X402. Now, the beauty of per call is you are not chasing 100 human subscribers, you're chasing one You're serving one capability that agents hit thousands of

10:08times a day. So, consider an example like YouTube transcription, right? If your agent is building this giant thing, if your software relies on this giant thing, um what's going to happen is it's going to ask um the internet, it's going to do research on how to transcribe a YouTube video, and if it finds an MCP, if it finds a an API that does that, it's not going to try and build a YouTube transcription tool itself. It is going to just use Do you see what I mean? So, agents are going to find the fastest and cheapest way to do things, and it is certainly cheaper and easier to use an

10:43existing API to transcribe those YouTube videos or to remove the background from those images or to do research or to do whatever, right? Than it is to build your own tool. So, you're serving one capability that agents need to do thousands of times a day while you sleep. Now, play two, idea three, is agent support or procurement, right? So, when the agent has a problem, where does it go? When a person's agent has a problem with a service, it needs somewhere to file a ticket, get a refund, follow up, etc. When a business agent needs to buy software, it wants to compare vendors against policies, get clean

11:20recommendations, pricing, etc. Both are wide-open gaps, right? So, you can build an agent support desk that wraps agent mail and a check-out protocol. Or you build a vendor comparison tool that agents consult before shortlisting. So either way you end up influencing what agents recommend to other agents. And remember that an agent is going to search for opinions on the best software. So AgentMail right now is one of the best email agent inbox tools, right? So there's not really much opinion. If you go and ask ChatGPT now, I need to build a software that has agentic email inboxes, right? It is going to choose AgentMail cuz it's just

12:00the most popular right now. You need to find a way with play two, idea three, which is the agent support procurement. You can build websites and directories that basically push [snorts] agents into a certain direction and then you can just shove affiliate links in there. Do you see what I mean? So an agent might literally go, find a tool that it likes, choose it, but because you you know, steered it

12:24in that direction, you make money. Now the other play is hiding in this shift. So any popular human tool reimagined for agents. So agent identity and permissions. So the one password and Okta for autonomous software. You've got receipts and audit trails. So what an agent saw, decided, changed, and bought. You've got security for agent inboxes. So you've got things like filtering, threat detection, um policy enforcement,

12:50that kind of thing, deliverability. Sandboxes for testing. So agents need staging environments to try integrations before paying. Then agent re- pricing pages as a service. So it's trivial to build, but it's broadly needed. Now the price you build is you or the price you choose is you price for a machine, not a person. So where founders from the human world get it wrong. So humans buy seats and monthly plans. An agent does not want a seat. An agent makes a huge number of tiny calls and wants to pay for exactly what it uses. So when my agent found the YouTube transcription API that I use, it uses that

13:31transcription API tool something like 50 times a day. And I never actually I was never actually in this loop. It just found it and chose it for me. So, you price per call, usage-based billing. So, keep a generous free tier. It matters more for agents than for humans because this it's how an agent discovers and trusts you in

13:49the first place, right? And the free tier is what you will be on before obviously the agent is building something that needs way more. So, an example is I only needed a couple of transcriptions a day, and now I pay for it because I need way more. And the math is if a call costs you a fraction of a cent and you charge a

14:06small multiple, every call is margin. So, machines generate calls at a scale humans never could. Now, importantly, AEO is the new SEO. So, I'm just talking about growth here. You can get cited, trusted, and recommended by AI systems. So, you can build the best agent tool in the world, but if no agent can find it, you have

14:26nothing. The old game of ranking on Google for human eyeballs is being replaced, right? So, the new game is the answer engine optimization. So, many sites, especially those buying Cloudflare, they now block AI crawlers by default, right? So, check that major ones can access yours. You add an LLMs.txt file, a short plain text file at the root of your site telling AI systems what you do. A non-technical founder can write a good one in an hour

14:52with help from Claude or what ChatGPT. Now, with AEO, you have to structure your content the way that AI reads it. So, you lead with a 50-word factual answer. That is the chunk of a chunk an AI grabs and quotes. Let's say you're using the YouTube transcription example. Excuse me. If you wanted to create an LLMs.txt or you wanted to write blogs, etc., all of my content would start with a fact saying, "YouTube transcript API that I've built um is the fastest and cheapest API transcription there. And obviously from there you write in clear factual statements. So agents extract

15:27facts, not vibes and not opinions. You add structured data labels. So prices, policies, offerings. You make machines stop guessing. You get listed where agents look, right? So public MCP registries, directories, marketplaces. They are being searched right this second by agents, right? And the end game is to be the tool that agents recommend to other agents, right? And it's just snowballs from there. And you're not worrying about humans recommending you anymore. You're just

15:54relying on the system. And that's the cheapest, most durable distribution there is. So to start this week, if you already sell something, spend this week exposing a simple MCP server for your software. You add an LLMs.txt file. You check your AI crawler access. And you create that LLMs.txt in a way that agents can

16:15understand. Or if you're starting fresh, you can pick one narrow capability that agent would pay for, right? Finding leads, scraping a LinkedIn page, finding a mobile number of a lead, doing research, transcribing a you know, transcribing YouTube video, transcribing a long piece of text, voice to text, whatever it may be. You expose it as a paid tool with a free starting tier. Then you list it where agents look. Whichever you choose, you build API first. You price per use. You resist the urge to spend 3 months on a beautiful interface because your

16:49customer does not have eyes. The bigger lesson is that the internet is splitting in two. There is the human internet we all grew up with. And now there is a new agentic internet being wired up right now. And agents are getting access to payments, identity, inboxes, and tools. Almost no one is building for the second one yet. The next wave of one-person businesses will be built by founders who saw the new customer arriving, spoke its language, and were ready to take its money. And everyone else was still optimizing for human who had already

17:18handed the decision to the machine. So, if you guys need additional help or you want the full setup of this, so you want to know how I build software and make money from it, the templates I use to build every project, the strategies I use to get customers, and 100 pages of info, go to rosewill.dev and I'll see

17:34you guys in the next video. Take care.