The AI mirage of the 90% solution
Originally posted on LinkedIn, February 10, 2025.
The feeling that a solution is almost there is one of the most frustrating and intoxicating parts of working with AI. It combines hope and despair at the same time.
At first, the progress feels incredible. A tool like Cursor or ChatGPT writes code and suggests solutions, and I think, “This is insane. I’ll finish it within an hour.”
Then one small detail fails. A line of code does not compile. An API does not connect. The solution feels “90% right,” but somehow it still does not work.
Instead of admitting that I have reached the point where the AI can no longer finish the job alone, I convince myself that one more attempt will fix it. It really does look that close, like a thread slipping out of my hand or a flickering light bulb that only needs one more turn.
Then the loop starts:
- One more question for the AI.
- One more carefully worded prompt.
- One small manual fix, followed by another test.
- One more line of code added or removed.
And time passes.
Instead of working on the problem itself, I work on getting the AI to understand me better. That is the breaking point. Instead of the AI working for me, I become an intermediary between myself and the tool, pulled into a game of wording and interpretation until I can barely remember what the solution was supposed to look like.
The closer I get, the harder it is to stop. If I were far from the solution, I would have walked away long ago. But I am so close. How can I give up now?
Sometimes the smart move is not to push one step further. It is to ask why I am stuck there in the first place.