A recent IBM Think newsletter article highlighted an intriguing milestone in artificial intelligence. While much of the world was watching the World Cup, an Anthropic employee used Claude to help disprove the Jacobian conjecture, a mathematical problem that had resisted proof since 1939.

One sentence from the accompanying discussion stayed with me.

“The model tried more things.”

I kept coming back to that sentence. The more I thought about it, the less it seemed to be about mathematics and the more it seemed to be about expertise itself.

Act I: The Economics of Search

For most of human history, solving difficult problems has been constrained by the cost of exploration. A mathematician could pursue only so many proof strategies. An engineer could prototype only so many designs. An architect could evaluate only so many alternatives before time, budgets, or simple mental fatigue forced a decision.

That limitation shaped more than the work itself. It shaped the way expertise evolved.

Experienced professionals learned to navigate enormous search spaces without consciously exploring every possibility. They recognized patterns, eliminated weak alternatives early, and focused their effort where experience suggested success was most likely. What we often call intuition is usually the accumulated result of thousands of earlier searches that have gradually been compressed into experience.

I’ve started thinking about expertise a little differently because of that.

Expertise is compressed search.

Act II: Following the Map

Once that idea occurred to me, several other observations suddenly fit together.

An experienced architect walking into a modernization effort doesn’t evaluate millions of possible approaches. Years of successes, failures, and difficult tradeoffs have already narrowed the realistic options to a handful worth exploring. The process feels instinctive because most of the searching has already happened.

Viewed this way, experience becomes something richer than accumulated knowledge. It becomes a remarkably efficient map of an otherwise overwhelming landscape. Experts aren’t valuable because they’ve memorized every answer. They’re valuable because they know where meaningful answers are most likely to be found.

That may also explain why experienced professionals often obtain better results from AI than someone using the same model for the first time. It’s tempting to attribute the difference to prompt engineering, but I suspect something more fundamental is happening.

Act III: AI Changes the Economics

Eventually I realized I wasn’t really thinking about large language models anymore. I was thinking about search methods.

It may be more useful to think of an LLM as a hierarchy of increasingly expensive search strategies.

Sometimes the statistical landscape strongly favors one response and an answer appears almost immediately. Sometimes several competing interpretations deserve consideration before one becomes more compelling than the others. When additional context is needed, the search expands through retrieval, external tools, code execution, web searches, or collaboration with a human. Each layer broadens the search while requiring additional resources.

The model is still exploring. but exploring at an unimaginable scale. That changes the economics in ways I don’t think we’ve fully appreciated yet.

Act IV: Judgment Becomes the Scarce Resource

For generations, professionals created much of their value by generating possibilities. AI makes generating possibilities dramatically less expensive. As that happens, judgment quietly becomes the scarce resource.

Organizations will have access to more ideas, more analyses, more designs, and more plausible solutions than ever before. The difficult question won’t be whether another answer exists. The difficult question will be deciding which answers deserve trust.

Architects have been solving that problem for a long time. Their value has never come from knowing every answer. It comes from defining meaningful constraints, exposing hidden dependencies, recognizing important tradeoffs, and deciding when enough exploration has taken place to move forward with confidence.

I don’t think AI diminishes the role of experienced professionals. I think it magnifies it. As search becomes less expensive, the ability to direct that search becomes more valuable.

Finale: The New Role of the Architect

Looking back, I wonder if that was the real significance of the sentence that first caught my attention.

“The model tried more things.”

The larger opportunity isn't simply allowing AI to search through more possibilities than we ever could. Deterministic and rules-based systems have been doing that for decades. The real opportunity is learning to shape that exploration toward questions worth asking in the first place.

Perhaps expertise has never really been about possessing the answers.

Maybe it’s always been about knowing where to look.