Let me cook!

Recipe making is how AI amplifies creativity

The world is drowning in AI announcements¹. But the real magic isn't in the next model drop—it's how you learn to combine what you already have.

The inspiration for this meditation comes from an under-explored feature of Claude Code². You can pipe operations on the command line into an LLM. Commandlinefu just got an upgrade.

cat file | claude -p "query" 

Don’t fret if you’re feeling lost, this is a convenient metaphor to stage the rest.

Recipe making unlocks real value

You and I have a long way to go to unlock the true amplifying power of LLMs and their applications.

I chalk it up to:

• Under-appreciating the Jagged Frontier. Solution: play with them more!

• Not taking full advantage of “piping”. Solution: play with them more!

Piping used here isn’t about exploiting “command line god actions” but rather knowing:

• What combinations of models and AI apps are good for which types of tasks (e.g. kitchen tools)

• What context or data you have available (e.g. ingredients)

• What task you have to complete (e.g. what you’re craving / in the mood for)

And then creatively combining them into conveniently delicious recipes.

Do external things distract you? Then make time for yourself to learn something worthwhile. — II. 7

Learning to Cook with AI: My Favorite Recipes

Song Production

• Ingredients: Conversation transcripts with my friends, trending music references, lyrical themes

• Tools: Claude for lyrics generation, Grok for musical structure, Suno for composition

• Process:

  1. Feed conversation highlights to Claude to extract text and emotional themes

  2. Use Grok to analyze trending songs and suggest musical structures

  3. Combine outputs into Suno to generate the complete song

  4. Iterate based on feedback

Software Composition

• Ingredients: Project requirements, user stories, technical constraints

• Tools: O1 Pro for PRDs, framework libraries, Claude Code for context, Cursor for targeted refinement, SpecStory for capturing my journey.

• Process:

  1. Generate comprehensive PRD with O1 Pro from high-level requirements that I personally

  2. Use framework libraries to scaffold initial project structure

  3. Leverage Claude Code for handling long-context design patterns and architecture

  4. Apply Cursor for targeted agentic refinement of specific code sections

  5. SpecStory to save and iterate through my prompts and thought process

  6. Iterate through feedback cycles to polish the implementation

Case in point: just last night I built this 3D snake game³ (that you can play online) using the above recipe in just 2 hours!

Course Content Creation

• Ingredients: Subject matter expertise, learning objectives, audience information

• Tools: Super Whisper for voice capture, Chorus.sh for multiple perspectives, Hemingway for simplification, my own brain for editing

• Process:

  1. Capture raw expertise through voice recording with Super Whisper

  2. Process content through multiple specialized LLMs via Chorus.sh

  3. Simplify language and improve readability with Hemingway

  4. Personalize and edit with my brain to maintain authentic teaching style

  5. Generate supporting materials for a complete course package

Case in point: this is exactly how

Hamel Husain

and I created what became ai-execs.com which turned into an O’Reilly book released on February 27th as AI Essentials for Tech Executives.

Building a talk

• Ingredients: Topic expertise, audience information, key points, time constraints

• Tools: Gemini 2.0 Flash Thinking for scripting, Gamma for slides, GIPHY+Claude for visuals, an “Augmented Speaker” for practice

• Process:

  1. Create conversational script with speaking cues using Gemini 2.0 Flash

  2. Build visual presentation with Gamma's AI-powered design

  3. Generate custom GIFs and visuals for key concepts using GIPHY API with Claude

  4. Rehearse with real-time feedback. I used ChatGPT’s voice mode for feedback when I was practicing!

  5. Record and enhance final presentation with Tella.tv

Case in point: this precisely how we then took the ai-execs.com material and cooked up a talk for the ai.engineer summit on February 19th.

The Next Frontier

As we march forward, the distinction between individual tools won’t be as important.

The future belongs to the chefs who figure out the best recipes that blend into seamless creative processes.

So don't just consume AI—cook with it.

Experiment with different recipes.

And share the unique outcomes that emerge.

1. Anthropic dropped Claude 3.7 and Claude Code dropped on February 24th. In attempt to not be upstaged, OpenAI released ChatGPT 4.5 just three days later.

2. Claude Code is a Terminal IDE.

3. New X trend is building single file ThreeJS games. Personal achievement unlocked. Code for the curious: https://github.com/gregce/snake_game.