--- title: "Agent Population Growth: A Funnel for Products Agents Choose — Ömer İlhan" description: "Agents are starting to research, try and pick products on our behalf. What it takes for a product to be discovered, used and recommended by them." url: https://omerilhan.xyz/blog/agent-population-growth-a-funnel-for-products-agents-choose author: "Ömer İlhan" published: 2026-10-15 updated: 2026-10-15 --- # Agent Population Growth: A Funnel for Products Agents Choose AI agents are now being used across different workflows. Sometimes they are used for web research, and sometimes for software development. Both developers and vibe coders are actively using them. Although agents may not be making purchases directly in every scenario yet, they are gradually moving toward a point where they can make purchases on behalf of both companies and individuals. As a result, the internet seems to be evolving into an environment where agents can navigate and take a wide range of actions. I want to illustrate this with a personal experience. ## A vendor selection without much effort I built a SaaS product and needed an email service to send emails to users during both sign-up and onboarding. As a vibe coder, I wasn't very familiar with email service providers on the technical side. So I explained my needs to an agent and asked it to research the available options. I wanted the integration process to be fast, the service to be easy to configure, the free tier to offer a generous usage allowance, and the solution to be cost-effective in the medium term. After researching the options against these requirements, the agent came back with a shortlist of email services. I then selected the option that seemed most suitable for me, Resend, and asked the agent how I could try it out quickly. I know plenty of vibe coders who ended up with Resend too. I created the account and the API key myself; after I added the key to my environment files, the agent handled the testing process. Once I got the result I wanted, I had effectively made a vendor selection on a small scale and could move forward with my project. > In other words, I had experienced agent-assisted vendor selection without putting in much effort. This could have been a database provider instead of an email service. And if we think about consumers, it could just as easily have been an e-commerce purchase. The only manual step was the account and the API key. That is exactly the kind of barrier agents still run into today. If a vendor is ready for the agentic internet, with a safe way for an agent to create and manage its own credentials, the whole process can have far less friction. Naturally, vendors that are easily discoverable by agents, offer an understandable experience, and make their products easy to try have an advantage. These qualities allow an agent to evaluate a product end to end and offer a suitable solution to the primary decision-maker: the developer or vibe coder. ## Two signals from Resend and Tally After experiencing this firsthand, I came across two posts on X that caught my attention as strong signals of this shift. The metrics shared by [Zeno Rocha](https://x.com/zenorocha) of Resend and [Marie Martens](https://x.com/MarieMartens) of Tally provide insights into how their services are being used through MCP and AI calls, showing significant growth in these usage categories. ![Zeno Rocha's post on X: 1 million MCP calls to Resend in August, up from 106,719 in April, their fastest growing usage category](https://omerilhan.xyz/blog/resend-mcp-calls.webp) *Resend: MCP calls, April to August* ![Marie Martens' post on X: Tally passed 1M AI tool calls in August, combining MCP calls and Tally AI assistant calls](https://omerilhan.xyz/blog/tally-ai-calls.webp) *Tally: MCP and AI assistant calls, April to August* Both companies seem to be positioning themselves for this emerging version of the internet. A note on these numbers: MCP calls measure how existing users work with a product through agents, not agent-driven adoption directly. Still, having assets like an MCP server and a machine-readable version of your product is what makes agent-driven adoption possible, and in my view it is a strong driver of it. Their documentation shows what this positioning looks like in practice. Imagine you've built a SaaS product. If you want agents to be able to use it, your MCP integration documentation should clearly explain the authentication process. Tally's [MCP documentation](https://developers.tally.so/api-reference/mcp) is a good example: it explains authentication in a few steps, with OAuth as the recommended path, shows the setup for each assistant, and walks through what an agent can do with example prompts. Its [API docs](https://developers.tally.so/api-reference/introduction) are written with the same clarity. Resend's [MCP documentation](https://resend.com/docs/mcp-server) goes a step further: a hosted server with OAuth, an API key path for headless agents running on a server or in CI, install instructions for every major client, and a full list of what the agent can do, organized by area, from sending emails to managing domains and even API keys, which is the credential barrier I ran into myself. Once the necessary authorization is in place, the agent should be able to access your product and use the tools you've made available. If you offer a free tier, the features included in that tier should also be usable by agents. An agent should be able to perform core functions such as creating a form with Tally or sending an email with Resend. When an agent can use your product, it also gets the opportunity to experience the value it provides. This means your product is no longer just an option that can be discovered; it becomes a solution that can actually be used within agent-driven workflows. ## Five steps to becoming a preferred solution So, what does it take to be ready for this shift? - Be discoverable. Agents need to find your product while they research. - Be remembered. Once discovered, your product needs to be reconsidered and recommended by agents during their research and evaluation. - Be usable. Make your product usable by agents and minimize the barriers they face during integration and access. - Be valuable. Once an agent can access your product, give it the tools and capabilities to experience the value your product provides. - Be preferred. If an agent can experience your product's value end to end, you've taken an important step toward becoming a preferred solution in the agentic internet. If this looks familiar, it is because it is the classic marketing funnel. What changes is who moves through it, and what convinces them at each stage. **The agent funnel** 1. **Be discoverable** (awareness): Found through search, docs and directories rather than ads. Machine-readable docs and an MCP server are the new landing page. 2. **Be remembered** (consideration): Shortlisted against the requirements the agent was given. Clear docs, pricing and a free tier earn a place on the list. 3. **Be usable** (trial): No sign-up form. The agent needs a safe way to authenticate, like OAuth or an API key it can create and manage itself. 4. **Be valuable** (activation): The free tier has to allow real work, like sending an email or creating a form, so the agent can prove the product works end to end. 5. **Be preferred** (retention): What worked once gets recommended next time. The agent carries the result into its next project and its next user. The Tally and Resend experiences above are good examples of these five stages. ## What changes for products and marketing Ultimately, as the agentic internet takes shape, marketing activities will likely need to adapt to how agents discover and evaluate products. Products themselves will also increasingly be designed not only around human usage patterns, but around how agents discover, integrate with, and use them. > Agent population growth will be the next growth cycle for those who are ready for it. I'll keep researching and learning about this topic. I'll explore the growth opportunities emerging from this new version of the internet in both the e-commerce and SaaS verticals and share my findings with you. See you next time, Ömer Tags: AI agents, MCP, Agentic internet, Product strategy, Go-to-market, B2B SaaS, Developer experience