I like tinkering with tools. I have a suspicion that it is an acceptable form of procrastination. Instead of digging the hole, I sharpened the shovel.
When you don’t feel like doing a task, build a tool for it. Write a script. Create a checklist of questions. Eventually, your brain starts selling itself product ideas instead of small utilities. You just need to give it some time.
It sounds strange, but sometimes this is enough to break through the wall of a difficult problem. Instead of trying to solve everything at once, you start making small improvements and gradually move forward.
Of course, this approach often leads nowhere. The experience is interesting, but the moment I try to explore even the first small piece, I often discover that I have once again invented a project that would take years to complete.
And not every idea, no matter how attractive, deserves years of investment. As a filter for finding projects that are both interesting and realistic, this approach works surprisingly well.
I think you can already guess what changed the rules of this little game.
Yes, of course: AI made it dramatically easier to build experimental tools and get initial feedback.
Recently, I had an idea to look into the state of WebMCP adoption. Naturally, it was part of another great plan™ that is not particularly important for this story. I just want to show the workflow.
WebMCP is a technology that allows websites to expose specialized assistants for AI tools to interact with their services. Instead of trying to infer capabilities from pages, forms, and widgets, AI systems can discover them directly. It is similar to a restaurant providing a menu instead of forcing you to inspect the kitchen to figure out what you can order.
Why do I think this technology matters?
Because the question of how to provide useful data and capabilities to your favorite AI agents is not something only I worry about. I have heard similar concerns from people working on browsers. The challenge is not simply converting a webpage into Markdown.
The real problems are pagination, content categorization, removing ads, filtering irrelevant comments, and figuring out which parts of a page actually matter. When a website can cooperate directly, life becomes much easier.
So, five minutes to describe what I wanted from the script. Twenty minutes of automated data collection. More than 300 popular domains were analyzed.
The result: only one of them had registered a WebMCP assistant.
To be fair, this was expected. The goal was to study early adoption attempts. Indexes like webmcp.com provide a good picture of smaller players who have already experimented with the technology. I was more interested in what happens among mainstream websites.
The result was not surprising. So I committed the research tool to my repository and set a calendar reminder to run the same analysis again in two months.
After all, every idea about building a tiny excavator deserves a chance to grow into a spaceport project.
Usually, this is where the story ends.
But notice something important: the total personal time investment was about ten minutes on a weekend morning. Five minutes to run the experiment. Five minutes to organize the results.
The point of this note is that you can do more. It is easy to find a few more ten-minute slots during the same day.
Let me tell you what I did next.
I took another tool out of my toy box.
It looked through the Chromium source code and prepared a report on what is happening with the Extensions API. And yes, there are already experiments in the browser codebase that could allow third-party sites to provide their own WebMCP tools.
That is genuinely encouraging.
The next step is to look at how this fits with the Terms of Service of popular services. Of course, things are not that simple. But if third-party WebMCP integrations become common, I suspect these restrictions will eventually have to evolve.
So now we wait.
The notes are written. The reminder for the next review is already there.
And there is still some spare time left.
For example, enough time to write this article.
This part I did without AI. After all, a hobby deserves some focused attention and properly spent time.
Partly because I want to enjoy the feeling that even the most ridiculous idea can earn its first ten minutes of existence.
If this kind of problem is on your desk — more about me, or get in touch. In the worst case, we will just have a topic to discuss, and each goes back to our own code.