I think you are right, and that tokens themselves can be a form of universal currency just like MCP+LLM forms a kind of universal glue between software. We tried to defend micropayments for content and support multiple times but worse alternatives always settled.

Maybe issues tagged with [tokens: 300k, accepted_models: opus-4.7-xhigh, pr_review: [opus>=4.6, gpt>5]] will become the norm. You provide the plan and prompt to orient the donator’s agent in your codebase, and they provide cost of implementation and review.

Since last week I strongly believe there will be tooling to apply an agentic conversation to a codebase, just like you can publish patches by email. Instead of providing the resulting PR, you provide the whole conversation that generates the PR, but the edits are there as inline patches and the whole transcript can be fed into a git CLI or GUI to see the diff. So you can’t masquerade a PR of sonnet 3.5 quality for an opus 4.7 one.

An interesting new kind of PoW.

Turns out we’re all in energy business (again).

How does it matter? Also, I am not using Claude for its miserable RAG, and proper solutions always involve several providers for different kinds of tasks.

I disagree we need to judge by the model used (it’s like judging the code quality by number of LoCs.) If I could have it done with local ollama—I’m good, I deserve a reward.

This is all not about donations, let alone control. I just want to be able to support my OSS better without breaking $1K/month limit on that, as I stated in the post above.