- I burned through around €100 of Sol and Fable in a few hours. The quality was good, but it is expensive.
- AI development sometimes feels pay-to-win. When I deliver better work, I have usually also spent more money.
- It often feels like I am treating people to tokens. Colleagues ask me questions for free, and sometimes I can literally forward those questions to an agent. LMGTFY.
- Agent loops get expensive quickly. Because of that, I run fewer loops than I would like. I can stay within budget by using Composer 2.5 to process incoming messages, but I would prefer GPT-5.6 Sol for that level of work.
- Today, AI was mostly about money.
Some positive notes:
- More and more often, I am learning from my agents instead of the other way around. That is one of the things I enjoy about working with increasingly capable models.
- I had Sol buy and configure a private Hetzner server. My laptop will become a thin client, while the Linux server does the heavy work. I am looking forward to the performance boost.
- I had some great brainstorming sessions with GPT Live. Conversations now feel surprisingly natural because the agent can respond in real time and occasionally take a moment to think by delegating to a smarter model. I cannot wait until it gets access to all my development and work tools, so I can spend less and less time behind a computer.