The Day I Realized I Was Asking the Wrong Question

I spent months asking “what can AI do?” One day I realized it was the wrong question.

If you, like me, spent the last two years reading headlines about how “AI now does Excel” or “AI now writes emails” or “AI now codes,” I have news: we’ve reached the end of the map. The exploration map, at least.

The phase of discovering basic capabilities is over. We already know AI can manage spreadsheets, create presentations, code, write reports, analyze data, automate emails, control smart homes, generate images, compose music, and debate philosophy. The catalog of “things AI does” is complete enough.

What’s not complete — not even close — is the catalog of “things I want to build using all of this.”

And that shift in question, so simple on the surface, is proving to be the most profound transformation in my relationship with technology.

Karpathy Hasn’t Written Code Since December

Andrej Karpathy — OpenAI co-founder, former Director of AI at Tesla, and probably the most-cited person on this blog — said something last week that crystallized everything for me.

In an interview on the No Priors podcast and later at Fortune Brainstorm Tech, he revealed he hasn’t written a single line of code since December 2025. AI agents handle about 80% of his coding tasks. He uses a “claw” called Dobby to control his entire house — sound, lighting, security, shades, HVAC, pool, and spa — all via WhatsApp messages in natural language.

But what really got me wasn’t the automation itself. It was his attitude toward the result:

AI sometimes doesn’t deliver perfect code, but he’s learning to let that go because it delivers the outcome he wants.

And there’s the paradigm shift I was slow to grasp. For tasks that “just need to work,” technical perfection is less important than agility and results. Karpathy isn’t optimizing code — he’s optimizing intention.

From Permission to Will

The key insight of this moment is that we’ve moved from the era of permission to the era of will.

We no longer need OpenAI or Google to tell us what’s possible. Almost everything that constitutes a reasonable business project has been “solved” by AI in terms of basic execution. The “vibe coding” tools market — the practice of describing what you want and letting AI generate it — reached $4.7 billion in 2025 and is projected to hit $12.3 billion by 2027. The Stack Overflow 2025 survey showed that 84% of developers already use or plan to use AI coding tools.

And Karpathy himself, who coined the term “vibe coding” in February 2025, has already declared it passé. In 2026, the right term is “agentic engineering”: humans no longer write most code. They direct, supervise, and orchestrate agents.

The question is no longer “what can AI do?” It’s: how do YOU want to shape AI to make YOUR life extraordinary?

The Two Pillars: Efficiency and Excellence

After months of experimenting, I’ve arrived at a division that works for me:

The Efficiency Pillar (“Just Work”). These are tasks where I let AI take near-total control. I accept a “good enough” result to save time. Scheduling, routine organization, routine emails, data cleaning, document formatting, repository monitoring. I describe the intention, AI executes, I confirm with a quick glance. I don’t care if the internal code is elegant. I care if it works.

The Excellence Pillar (“Human Taste”). This is where I put my finger, my judgment, and my quality criteria. Strategic analysis. Communication that needs authentic voice and tone. Decisions requiring ethics and context that AI doesn’t have. Content creation where I want it to reflect my perspective, not a statistical average. This is where I create something AI alone couldn’t.

The temptation is to put everything in the first pillar. But when everything is “good enough,” nothing is extraordinary. Human taste is what separates the mediocre from the memorable.

From Chatbot to Tool Creation

A practical change I made that I strongly recommend: stop using AI only as a chatbot. Start using it to build tools.

If you have a recurring problem — like organizing shift schedules for 6 people, or consolidating weekly reports from multiple sources, or monitoring competitor pricing — there are two paths:

The basic level is throwing data into a chat and asking for the answer every week. It works, but it’s manual and doesn’t scale.

The advanced level is using Claude, ChatGPT, or any agent to build a mini-app that automates it forever. A tool that runs on its own, with defined inputs and structured outputs.

And here’s the transformative point: even if you’re not technical, this is now possible. Vibe coding tools — Lovable, Bolt.new, Replit for prototyping; Cursor, OpenClaw, Claude Code for development — let you describe your intention and AI writes the code for you. Non-technical people have built products generating real $1M revenue using Claude Code.

The power to create software has left the exclusive hands of programmers and moved to those who ask the best questions.

The “State of Psychosis” of Those Paying Attention

Karpathy used an expression that hit me hard: he said he’s in a “state of psychosis” trying to figure out what’s possible and pushing the limits. And he thinks this psychological state is common among frontier researchers right now.

I relate. Not at his level, obviously — I’m not an OpenAI co-founder. But that feeling that the field is moving so fast that if I blink, I’ll miss an enormous opportunity… that one I know. And I bet you do too.

The antidote that worked for me was to stop trying to keep up with everything and start focusing on what I want to build. Not what’s possible in the abstract, but what solves a real problem in my life or my work. The list of possibilities is infinite. The list of things that actually matter to me is much smaller — and much more actionable.

Conclusion: Human Taste Is the New Currency

In a world where AI can do almost anything, your good taste, your judgment, and your vision become the most expensive assets on the market.

Use AI for the mundane, so you have time to be excellent at what truly matters. Don’t drown trying to master every new tool. Choose your pillars. Define where you accept “good enough” and where you demand excellence. And build tools — not just answers.

The era of asking “what can AI do?” was 2025. The question of 2026 is: what do I want to build?

And that question, no AI can answer for you.

Share if this resonated:

The power to create software has changed hands. It moved from those who know how to write code to those who know how to ask the right questions.

Read Also