AI Only Executes What We Imagine — Why Deep Thinking Beats Execution Now

AI can only execute work that someone has imagined.

In Software Engineering, imagination often comes from thousands of hours of expertise: you can't imagine with capabilities you haven't touched; you can't ask deep questions based on just a summary. We think that SWE is done because AI can solve any task now. This is true, but the question is: define "any task". Any task means "any task we used to have". So the accurate statement is: AI can solve most of the tasks SWEs have been solving before AI. And yes, solved problems don't bring value.

The Shift to Impossible Problems

Value has shifted towards "impossible problems". Problems that were considered extremely hard or not worth pursuing given the estimated low ROI. Problems where, just by trying to solve them, we could have solved solvable problems that bring real value.

If your to-do list before AI is the same as yours after AI, you are selling bread to bakers. Your new to-do list has to be a set of crazy, divergent, challenging, and value-oriented ideas that only AI, not humans, is ready to tackle. Then you enter the next-level battlefield, and then you solve problems that not everyone can solve now.

The Death of Execution as a Competitive Skill

Before AI, repeat-till-mastery was a competitive skill in itself that, even without much creativity, could grant you a comfortable position in the SWE race: you could execute complex and creative ideas that shouldn't necessarily come from you personally. Now, the execution part is progressively shifting to the AI side and asymptotically converging to what top human talent used to have. This means creativity and deep thinking are the key differentiators, more than any time before.

Three Paths to Creative Excellence

How to gain this new golden asset? There are at least three paths:

1. Deep Hands-On Expertise

The first is what most senior+ SWEs gained through the years: deep hands-on expertise built through real-life problem-solving. This fills your mind with moments when you wished there was a better way to do things, but realized that was going to be too challenging to try.

2. Academic Background and Research

Second, a deep academic background. We often considered researchers as dreaming utopians that overcomplicate problems and find solutions to what does not exist before it exists. Today, many of them have what they were too busy to gain: execution! Many papers and many theories that were placed on the shelves just because execution was too expensive to justify the costs are now more than welcome to be wired together to prototype and see if they can be the perfect solution, instead of the legacy ones we kept using because trying something else wasn't cheap.

3. Neurodiversity as an Asset

The last path isn't actually a path, but just a gift. Some people are neurodivergent by nature. They have been either forced by their surroundings or forcing themselves to fit the process because that was the way to align with the big machine that brings value. Today, they have an incredible chance to validate many of their crazy ideas and prove their odd path was really the best path.

AI Slop is a Human Problem

The quality of AI output depends entirely on the quality of human imagination and intent behind it. AI slop is not an AI problem—it's a reflection of lazy thinking, poor prompting, and lack of deep understanding of what problem we're actually solving.