Listen instead

This material reflects my opinions and not those of my employers.

In my last post, I mentioned I wanted to report on how the automated blog publishing process actually went.

The short version: it mostly worked, but again, the interaction with AI required good understanding of technical terms and technology. While eight out of ten steps ran without my interaction, there were a couple it requested me to step in and complete them. And as I suggested in part 1, the instructions would not allow my mother to complete it by herself!

For example, here are a couple of excerpts from the conversation that would be a blocker for a non-technical person:

The sandbox is blocking outbound API calls with a proxy restriction — this is a network limitation on my end, not an issue with your credentials or the script. The ElevenLabs call gets a 403 Forbidden from the proxy before it even reaches ElevenLabs.

Here’s the cleanest workaround: I’ll save the exact audio text to a file, and you run the existing script on your Mac with a small tweak to point at that text. It’s a one-command step.

Waiting on you: Run the TTS script from your Mac inside code_folder/:

source venv/bin/activate
python saas_tts.py

Caveat: I did add some slightly technical terms in my prompt and it may have inferred it could speak tech with me. It may have responded in a different way for a completely non-technical prompt. Something I will try in the future.

Positives

It saved me something between 1 to 2 hours. That is huge for what is a relatively simple task for me. It definitely shows the potential to save me a lot of time in repetitive technical bureaucracy. The question goes back to my previous post “The Cost of Thinking” about the financial trade offs. For me, it personally pays off because the entire set of tasks fit within my $20 subscription, but what was the real AI thinking costs to do it compared to using my brain for 1.5 hours?

Well, it turns out, based on non-exhaustive review, it would be more expensive in this case to burn my calories working on it by myself, particularly considering the small number of tokens required and what the hourly cost for a human professional could have been. The difference is that instead of paying a human to do it, I am paying for compute and energy.

The verdict

Of the ten planned steps, I was hands-on for three: running the TTS script on my Mac, pushing commits to GitHub, and merging the PR. The merge is by design as I want to review before anything goes to production.

I will definitely use AI for blog publishing going forward. It gives me personal time to do some other things, perhaps teaching my mother to use AI!