AIMM Session — July 31, 2025: The Prompt Is Dead, Long Live the Process

“If you want consistent results, build reproducibility — not one-off wins.” — Lou

30-Second Summary

Lou went deep on the idea that prompting as a skill is already obsolete. What replaces it isn’t better prompts — it’s process architecture: systems that dynamically engineer their own context, generate their own prompts, and adapt in real time. The group watched Lou build an interactive Claude artifact live in under an hour — one that can be handed to a client as a lead-gen tool, no coding required. Guest Kee Jones compared it to a master’s degree. Donald summed it up: “That was AI waterboarding.” Elizabeth: “Every time you get carried away is a present to us.”

1. Stop Writing Prompts. Start Building Pipelines.

The bucket vs. the pipeline. Three levels of abstraction:

Level 1 — Task prompt: Does the actual work.
Level 2 — Meta prompt: Generates the task prompt dynamically. Domain-aware.
Level 3 — Meta-meta prompt: Domain-agnostic. Interviews the user, understands the goal, generates the meta prompt.

The interview with the prompt happens at the thought level, not the execution level. Don Back: “Systems thinking rather than task thinking.” Donald Kihenja: “Abstraction.”

2. Context Engineering Is the New Prompting

“Priming” the prompt now has a formal name: context engineering. You encode the priming strategy into the prompt itself. Lou demonstrated with a real-time article on “process over prompts” — simulated a debate between two expert personas (Dr. Ada Vance vs. Julian Hart), declared a winner, used the enriched context to generate a 1,500-word article in 10 minutes. Don Back: “This gets beyond the surface-level stuff polluting the internet.”

Key phrase: Reflect, don’t act. Tell the AI to think about what it’s going to do before doing it.

3. The AI Tutor That Grades Itself

A prompt blueprint from Nathan B. Jones — a personal AI tutor that only advances you when you score 80% or better on the current lesson. Diagnoses starting point, teaches in 250-word micro-blocks, gives practice challenges, evaluates. Don Back: “A glimpse of the future of learning.”

The magic line: “escalate difficulty only when the student shows 80% competency.”

4. From Legal Jargon to Mediation Brief: Process in the Wild

A construction company client had a legal AI app but got vague results. Lou’s fix — not a better prompt, a better process:

Define a persona: top 1% legal researcher, “Sherlock Holmes for depositions”
Run a query rewrite for semantically precise retrieval
Execute multi-query expansion — not just what was asked, but what’s needed to answer it
Surface 100+ actionable items from the deposition
Follow suggested threads: find weaknesses, find defenses, compare testimony
End with a thorough mediation brief

“There was nothing different about the system at all. All it was was knowing how to pull the right information into the context.”

5. The Interactive Artifact: From Idea to Tool in One Session

Live-built a Claude artifact — a fully interactive prompt design partner that:

Asks for topic, audience, and constraints
Analyzes the intersection
Generates a contextually optimized prompt
Lets the user review and edit before running
Runs the prompt and displays output
Includes a prompt effectiveness self-evaluation

Deployment options: share the Claude artifact link, or export as TSX to Netlify, Vercel, Replit, DigitalOcean. Donald Kihenja: “Firebase Studio will create the code, deploy it, and give you a free URL.”

Hot Takes & Resources

Model pairing: O3/O4 for brainstorming; Grok 3 for thinking faster; Claude for writing with custom styles
PromptLayer’s Prompt Blueprints docs: https://docs.promptlayer.com/running-requests/prompt-blueprints
Donald’s affirmations app (Firebase Studio): https://studio—zentube-affirmations.us-central1.hosted.app/
“You don’t start by creating the Custom GPT. You start at the thought level — at the strategy level — and then have two layers of prompt generators create the ultimate prompt for you.”

Community Corner

Don Back is building a 12-email nurture sequence and had a light-bulb: “I’m going to step back, look at this as a process, and build the prompt that builds the prompt that builds the generator.”

Jamie W generating output “that reads like an MIT dissertation.” Challenge: calibrating audience-fit.

Donald Kihenja demoed Firebase Studio as full-stack free hosting.

Kee Jones (guest via Bally): “It feels like getting a master’s degree.”