In the Age of AI, I Want to Be Closer to Real Problems

— We no longer lack the ability to solve problems. What we lack are real problems to solve.

The Impact

Over the past month, Claude Code has hit me hard — even harder than when I first saw ChatGPT at the end of 2022. When Claude Code opened Chrome to verify its own changes and told me everything passed, I felt the same rush I got years ago from using a WoW bot (HonorBuddy) to fish out a turtle mount. That feeling of using automation to bypass a tedious, repetitive process — it’s fascinating.

Tool Fever

Following the momentum, I spent two weeks vibing obsessively, trying to automate everything I do at work. The whole company was in the same frenzy — skills, plugins, agents, raining down like hail. Then, after all the dizziness, I stopped and asked myself: where is my turtle mount?

With Claude Code, building tools became just easy — and also much less important. Tools without real-world validation often just burn through context for nothing. Any productivity gain has to eventually show up as solving an actual problem. I stopped the automation craze and turned my attention back to the projects at hand.

Distilling to the Void

I finally returned to work itself and let Claude Code take over all my projects. It built structured memory; it slowly absorbed domain knowledge and workflows from our conversations; over two weeks it grew fast, and productivity improved steadily. But then I hit another wall. I call it distilling to the void: when you distill your work through Claude Code, what you end up with is just a manual for one tiny gear inside a giant machine.

I skimmed through those Markdown files. They were neat and abstract. Mostly: how to use the internal tools, how to find another gear in this complex system, how to find the local optimum for this small team within a massive optimization equation. Plus some scattered code style notes — the kind you can find anywhere in open source.

I know that’s just how it is at big companies. You keep your gear spinning, and occasionally deal with the friction of org-level politics. But Claude Code made me see it more directly. That’s disappointing.

Too Far From the Problem

What actually demoralized me wasn’t realizing I’m just a cog. It was realizing I’m too far from real problems. By real problems, I mean needs that are directly tied to real-world value. Take my work as an example: I’m in digital advertising, specifically building features that let advertisers run ads to WhatsApp users. This matters to the company — it drives a lot of revenue. But it also doesn’t really matter, because outside of Meta’s platform, it doesn’t solve anything. So I don’t feel like I’m working on a real need.

I pushed back on myself from a few angles: On the advertiser side — do you actually know how to run campaigns in a way that delivers the most value for different industries? On the user side — do you know what kinds of ads resonate with users in different countries? On the technical side — do you understand how this massive system stays stable?

These real problems have already been broken down, dissolved into smaller pieces inside the big machine, and cleanly solved. As a small cog, I know what I’m doing — I just don’t know why it matters.

Closing

Without AI, I think everything above would just be a complaint. But now, Claude Code and other AI tools have given everyone essentially unlimited productivity. Being far from real problems is no longer just unfortunate — it’s a fatal flaw. Productivity with no real problem to absorb it just burns away like wasted tokens.

Whether for companies or individuals, the future will belong to those who can find real problems.

In the age of AI, I want to be closer to the problem.