AI Gives Standard Answers. Humans Find Nonstandard Solutions.

I was using Codex to write a program today—em... to be precise, to write the development plan first—and came away with a thought.

If I had to sum it up in one sentence: AI is better at giving standard answers; humans are better at finding “nonstandard” solutions.

AI, or Codex, is like a meticulous engineer. It cares a great deal about “engineering correctness,” or perhaps “engineering precision.”

Put simply, it will insist on precision in certain details and devise more complex engineering solutions to achieve its goal, making the program as reliable as possible.

There’s nothing wrong with that, really. I just feel that sometimes it wastes time, tokens, and resources—for example, by introducing several dependencies or adding multiple intermediate steps.

Watching it work through the problem, I kept thinking: there’s clearly a simpler, more efficient way. Why get so stuck on this? Though it does look professional and dedicated.

Then something clicked. AI is better at giving standard answers, perhaps because the knowledge and data it was trained on come from the average of a vast amount of human knowledge and experience. Maybe a little above average. But broadly speaking, it’s the average, so AI tends to produce standard answers.

Human society is so varied, though—how often are things really black and white? In many situations, there are “nonstandard” answers. There is room to be flexible.

For example, I knew there was a simpler solution here because I know 剪映 (Jianying, a video-editing app) well enough to know which features can compensate for flaws in a recording and provide a fallback. And beyond fixing things at the editing stage, I could also adjust the recording workflow to prevent some of the problems in the first place.

In other words, for most problems, if technology alone doesn’t offer a way through, combining technology with a different management approach or a redesigned business process can still achieve the same goal.

To my mind, all roads lead to Rome. You don’t have to keep hammering away at the engineering. But AI doesn’t know that, so it diligently burrows into the details.

People can offer “nonstandard” answers because hands-on experience across many real business situations gradually develops an intuition for solving problems.

That intuition isn’t in AI’s training data. It lives in human experts’ tacit experience and industry know-how.

This is also the most valuable part of what FDEs—forward-deployed engineers—offer today: using know-how in a specialized field to give clients professional solutions that go beyond AI’s standard answers. Those solutions are very likely to be “nonstandard” and customized.

I suddenly remembered working at an office SaaS company years ago. Sometimes I would give clients a dazzling presentation of a proposed solution, while knowing perfectly well that perhaps more than half the features weren’t in our standard product.

I simply knew very clearly what the clients wanted and what they wanted to hear, so the PowerPoint presentation would speak directly to them. Whether we actually built it was a conversation for later. There was plenty of room to work things out.

You see, in those days, SaaS vendors were fixated on building one standardized product, selling it to ten thousand or a million customers, and sitting back to collect the money. Yet every client you visited wanted something tailored to them, something unique. That’s one of the reasons I think making SaaS work in China is so difficult, perhaps even unworkable.

But!

In the AI era, with a technology that has personalization and flexibility in its very DNA, suddenly anything seems possible.

(P.S. A little detour: hasn’t 梁圣—an admiring nickname for DeepSeek founder 梁文锋, Liang Wenfeng—already laid the groundwork for us? Everything is plugin. Sit with that for a moment. Really think about it. So, free your mind~)

Back to the point~

AI’s preference for standard answers isn’t a fault; it’s a characteristic. By all means, use it for problems and work you’re particularly bad at. At least it can get you to a passing 60 out of 100.

But in the areas where you truly excel, make good use of your tacit knowledge, experience, and business intuition: those life-trained parameters you’ve forgotten are tucked away somewhere in your brain or memory. A particular situation may activate them. Then, working with AI, you can guide it toward better products and solutions. That value bears your name.

Seen this way, perhaps we should all have more experiences, whether good or bad. Aren’t those the most precious things that truly belong to us?

All right, I’ll leave it there. Let your imagination take it from here.

Good night~