AI agents produce work faster. But do they actually reduce it?
We write less code and copy, then spend more time reviewing unfamiliar output, resolving contradictions, and deciding what is safe to ship. Sometimes that is leverage. Sometimes it is review debt.
The real test is not how much an agent generates, but how much work it removes end to end.
Where has AI genuinely taken work off your plate and where has it merely moved the work?
AI now lets one person build far beyond what they could write or understand line by line. Some see that as the natural next step after libraries, frameworks, and open-source dependencies. Others think shipping code you cannot explain means giving up control while keeping all the responsibility.
Are passing tests and a successful review enough? Or should a programmer be able to understand and debug every critical path before it reaches users?
An AI can reproduce mechanics, generate a polished prototype, and pass every automated test. But it has never felt tension before a boss fight, relief after surviving with one hit point, or the satisfaction of finally mastering a difficult movement. That raises a question I keep coming back to: is a good game built from patterns, or from taste? We can measure whether a build launches, inputs work, objectives are reachable, and old behavior still passes. The harder questions are subjective: Does movement feel satisfying? Does the pacing drag? Is the difficulty fair? Is the core loop still interesting after ten minutes? Maybe AI does not need to feel fun if it can observe players retries, hesitation, quit points, input patterns and iterate under human creative direction. Or maybe those signals only optimize engagement and can never replace taste. Where would you draw the line?
