More AI Thoughts (Part 2)
Notes from May: Cultivate Taste as an Explicit Asset
The first half of 2025 has been a blur.
Each month, a new model. Each week, a new launch. The feeds are flooded. Keeping up feels like trying to sip from a firehose.
Some ideas I had written down months ago:
“Accept, Accept, Accept” → now Vibe Coding.
“Spec-to-Product” → now Prompt sets are the new PRDs.
Below are some new thoughts (Read Part 1):
1 Taste is the differentiator
I used to think clear thinking and goal-orientation were the foundations of great AI work. That’s still true—but incomplete.The real game is taste.
Taste is the ability to know what you want, describe it vividly, and recognize when it’s good. And taste isn’t just intuition—it’s built on cognition and knowledge.
Wittgenstein said, “The limits of my language are the limits of my world.”
The limits of my taste are the limits of my AI.
I can’t ask AI to make what I can’t mentally see.
For instance, I only know to ask GPT-4o to create images in Studio Ghibli or anime styles—not because I love them, but because I don’t know more.
2 Embrace the abundant mindset
Treat AI as infinite intelligence.
Ask. Debate. Iterate. Obsessively.
Don’t stop at the first good-enough response. Regenerate.
Vary phrasing. Switch models. Ask sideways questions. More prompts, more angles, more ideas.
You’ll be surprised how different the fifth answer can be from the first.
Diminishing returns do kick in eventually. The trick is knowing when to stop. And that’s… taste again.
3 Context and memory are underrated superpowers
I saw a Product Manager write, “Just throw the sales data to AI and it’ll give the insights to the users.” That’s not a product spec—it’s wishful thinking.
Context matters. System prompts matter.
Prompting is a form of compression. Design for emergence, not just extraction.
Think of the Tao Te Ching: a few hundred words compressing infinite wisdom.
True memory is structured. It’s semantic. It evolves.
RAG is part of it. But so is your shared context with the model—the accumulation of conversations, feedback, and course corrections.
The more you talk to an AI, the more likely it syncs with your patterns.
The memory isn’t just what it knows. It’s who it’s learning.
4 Everyone is a manager now
Peter Drucker said, "Knowledge workers must, effectively, be their own chief executive officers."
In the AI era, every one of us is not just managing oneself, but also managing teams of AI agents.
Prompting isn’t just instructing. It’s delegation. It’s shaping. It’s leading.
And yes, that means agency matters more than ever.
The people who thrive? They direct. Decide. Reflect.
Final note
Some treat it as a search engine. Some treat it as an operating system.
Which one are you?