The Token Manifesto

We are uncovering better ways of consuming context by doing it and helping others do it.
Through the work of producing tokens, deleting tokens, and staring at a usage bar nervously, we have come to value:
1. Short system prompts over clever system prompts : The model reads the same 47 lines you wrote at 2am and judges you either way.
2. One clear example over three paragraphs of explanation Show, don't tell. Show, don't tell again.
3. Iterating in small steps over dumping the whole spec at once : A 200-line brief is not "rich context", it's a cry for help.
4. Outputting in a defined format over letting the model freestyle : If you wanted prose, you wouldn't have asked for JSON.

That is, while there is value in the items on the right, we value the items on the left more.

The 12 Principles of Token Efficiency

 1. Our highest priority is to satisfy the user through early and continuous delivery of the answer, not the lead-up to the answer.
 2. Welcome changing requirements, even late in the conversation. Especially late in the conversation, actually, at that point you're committed.
 3. Deliver working prompts frequently, from a couple of weeks to a couple of minutes, with a preference to the shorter timescale ; not because we like rushing, but because you're paying.
 4. Business people and developers must work together daily throughout the project. Also, they must both work with the AI. Also, the AI is doing most of the work.
 5. Build projects around motivated individuals. Give them the environment and the context window they need, and trust them to get the job done.
 6. The most efficient and effective method of conveying information to and within a team is face-to-face conversation. The most efficient method of conveying information to a model is one well-chosen example and a single "like that, but…".
 7. Working software is the primary measure of progress. A working prompt is the primary measure of your credit bill.
 8. Sustainable prompting — the sponsors, developers, and users should be able to maintain a constant pace indefinitely. If your prompt is 4,000 tokens, you are not maintaining a pace, you are training for a marathon.
 9. Continuous attention to prompt clarity enhances response agility. Definite prompt, indefinite answer.
10. Simplicity, the art of maximizing the amount of work not done by the model, is essential.
11. The best architectures, requirements, and prompts emerge from self-organizing teams who notice they're repeating themselves and write it down once.
12. At regular intervals, the team reflects on how to become more effective, then tunes and adjusts its behavior accordingly. Especially before the monthly invoice arrives.

"You don't have a prompt problem. You have a context-window problem. And a writing problem. But mostly the first one."

also

"Everyone's a prompt engineer until they run out of monthly quota."

Content created by Nicolas Martignole, GLM-5.2 and MiniMax-M3