Plutonic Rainbows

Plutonic Rainbows

Spiralling Costs

Coding with agents is becoming expensive, and it’s surprisingly easy to accumulate high costs. To reduce token usage, I’m working on creating smaller, more specific prompts. Careful planning and a clear structure are also essential.

Prompts

I created my own prompt generator, trained with OpenAI, after growing frustrated with Claude Code and Cursor misinterpreting my prompts.

Monday

I had Sonnet 3.7 generate README and CHANGELOG files for all my projects. I suppose relieving this kind of monotony is a worthwhile use of agents.

OpenAI released 4.1 today with variants — all available through their API.

Agent Woes

Using Agents to write your code is likely the future. Today, I used them to add icons to a menu bar in an image generation app, though it wasn’t entirely smooth sailing. Be very careful — and, more importantly, ensure you phrase your prompts correctly. This will save you from countless hours of panic, rewrites, and GitHub reverts.

Anyway, managed to get the icons working properly and fixed an issue where the endpoint would not switch smoothly. Yesterday, I also added progress indicators.

Money Wasted

I spent unnecessary money trying to get Gemini 2.5 Pro Max and Sonnet 3.7 Max to create file attributes for projects and apply them to a new folder. Two things came of it:

  • Premium tool calls — $15 in total. Totally wasted money.

  • I made no real progress and ultimately just created a boilerplate template instead.

  • The boilerplate worked with only minimal edits.

The only positive from the experience is that I managed to completely revamp the UX for my image generation projects.