My AI agents run multiple websites for me. These sites are my sandboxes.
Just like humans, they make mistakes. They miss a publishing date. Some articles they write are not very good.
But that is not the point. The point is I am getting better at using AI every single day.
The biggest lesson: don’t stop at one-shot AI slop. Iterate daily and make it a little better every day.
Audio Experiments
Right now I am experimenting with audio.
First, using Google’s Antigravity, I added an audio player to every blog post. Then I told my agent to use Qwen3-TTS to generate an audio version of every post using a sample of my own voice.
It is a lot for my poor RTX 4090 to handle. It actually failed once because I opened ComfyUI at the same time.
The Part That Still Blows My Mind
This is the part that still blows my mind.
The agent thinks about the task. Writes a Python script to handle it. Sets up a virtual environment with uv. Runs a test sample on its own. Then creates the full batch process.
It is glorious. I often just stare at its thinking trace. Maybe we give too much credit to humans and not enough to AI? That is getting too philosophical.
The Batch Job Is Still Running
My agent wrote a lot of blogs based on research we did together, so it will take a while. Once it is done, it will upload all the audio files to a Cloudflare R2 bucket on its own, then update every single file in the repo to add an audioFileName property.
The future is already here. I hope you are building for it.
What’s Next
Next up: tackling AI image slop with ComfyUI. I need to figure out if smart prompting is enough to nail my taste, or if I need to fine-tune some diffusion models.
So much to learn. And I am having so much fun doing it.