I asked Henk (my OpenClaw bot) to look at all my experiments and come up with a new one it could run itself.

Can you start investigating all of the experiments done on hugomelis.nl and come up with a new AI experiment to run yourself? You are allowed to then do the actual experiment as long as it can be done in max. 15 minutes.

It came with the following response:

I picked a promising new experiment direction: “Can my own experiment archive tell me what people would actually pay me for?” Why this one:

  • it fits the existing theme of the site
  • it can be run in under 15 minutes
  • it produces something useful for the business side of hugomelis.nl

OK, interesting. I am currently thinking of adding a service so I can help other founders and executives implementing AI agents and tools in their life and work.

So I asked it to run the experiment. It started working and it came back with the results after about 13 minutes. It went through all of the experiments on the website and it said it extracted the voice of the posts.

It then wrote the complete experiment post for me in my own voice. But this is where I still feel there is so much room for improvement. AI text is just a perfect combination of sentences, something real text isn’t most of the time. And AI text is often dull to read. Even though it had a lot of earlier content to base its voice on, so it would be similar to my writing, it didn’t get it.

Let me show you some examples of what it wrote:

I started these daily AI experiments because I did not want to become obsolete after selling my app development agency. At first they felt like a way to keep up: touch the tools, document what I learn, stay close to what is changing. But after enough experiments, another question starts to show up.

If someone reads this archive from the outside, what do they think I actually do? Not what I say I do. Not what I plan to offer later. What does the body of work itself suggest?

So for this experiment I looked through the published experiments on hugomelis.nl as if I were a stranger trying to infer the business behind the writing. The answer was less subtle than I expected. So the archive does not read like “man tests AI tools.” It reads more like “man helps people turn AI into something operational.” That is a much more commercial statement than I usually allow myself to make directly.

Really this is not my own voice, it just sounds too perfect. Let’s take: “if I were a stranger trying to infer the business behind the writing.” or “That is a much more commercial statement than I usually allow myself to make directly.” for example. It sounds too profound for me it feels.

In my experiment write ups I get coached by AI to make sure the story arc is there, and I be specific in showing examples, things like that.

The main things is that I am not fixing all of the suggestions, not because I think they aren’t valid, but more because I still want it to sound like me.

Not everything was bad though. As Henk found an interesting observation when analyzing the experiments: “man helps people turn AI into something operational.”

Yes, this is actually what my mission is. Help people implement AI tools in their daily life, so that they won’t become obsolete. I just need to figure out a way to make this into a service or product when the time is right.

Key Insight

The coaching helps with the story arc and specificity, but if I accept every suggestion, I still end up with a text that reads like Henk wrote it. The suggestions are usually correct, but “correct” and “mine” aren’t the same thing.