Beyond Prompting: From Super Prompts to Process Prompts

The frame

Lou opened with a question he keeps returning to: what is the 10% of effort that produces 90% of the result for a knowledge entrepreneur working with AI? His answer, said plainly, was prompting. But the version of prompting most people are doing is the version that wears them out, copy-pasting templates, iterating turn by turn, building libraries of brittle one-offs.

He walked the room through the evolution he sees:

1. Conversational prompting. Just talking to the model.
2. Structured prompts. PROMPTS frameworks, persona, role, output, methodology, terms, shots.
3. Super prompts and custom assistants. Captured sequences with embedded context.
4. Meta prompts. Prompts that generate the structured prompts.
5. Process prompts. What Lou is now coining as the next layer, prompts that interview the user, infer the variables, plan the work, and then execute a generalized framework.

What landed

The shift Lou wanted the cohort to feel was the move from writing prompts for a specific use case to designing prompts that produce reusable assets. The article you write today should not be the work product, the prompt that wrote the article should be the work product. Run that prompt next week with different inputs and the system still works.

He demoed this against Claude. Instead of giving the model a single article brief, he had it interview him about his frameworks, infer the audience, propose the structure, and only then produce the draft. The cohort watched the model do something that felt closer to thinking with him than executing for him.

The chat side conversation

Don Back wrote in the chat that this was deeper than anything he had been able to specify for a contract writer. Bally said she was trying to keep up. Donald named what was happening: building custom thinking machines for multiple uses rather than trying to build something one of. That phrase stuck for the rest of the call.

What the cohort took away

The action that emerged was simple and not easy. Take one repeated piece of work in your week, the kind you have prompted twenty different ways across twenty different chats, and write the process prompt for it once. Use the AI to write that prompt. Then never write the underlying prompt again.

The closing note

Lou ended where he started. The skill is not prompt writing. The skill is knowing what to ask, in what order, with what constraints, so the model produces an asset that does the work for you. Everything else compounds from there.