What AI Can't Do for You

I’ve been on the internet long enough to recognize when something crosses the line from hype into the next big thing. When ChatGPT launched in late 2022, I got on it and thought, yep, this is going to be interesting. Not out of fear, but out of genuine curiosity about where it was headed. After skimming the white papers and listening to what industry leaders were projecting, one thought kept coming back: this was going to let a lot of people generate a lot of things.

My background isn’t deeply technical. I’ve spent years in a direct support role next to middle management, and that seat gave me a clear view of what was coming. Years ago at a development studio, the owner had a phrase that stuck with me: “the race to the middle.” It was what happens when there is an abundance of the same thing, and that glut drags down the value of everything around it. Back then, LLMs could barely produce anything worth keeping. But I knew that would change. Fast forward to 2026, and we are in the race.

I was reminded of this recently at an AI event here in Costa Rica. The talks circled the topic without ever really landing on it. What I actually saw in the room was fear. A lot of the people there were technical, builders and developers like me, and they are quietly scared that AI is going to do everything, and that there won’t be much left for the rest of us. I understand that fear, and I take it seriously, because the people carrying it are not naive. They are watching something move fast. But the fear is pointed at the wrong thing, and I want to walk through why.

The Race to the Middle

Here is what the race to the middle actually looks like. More and more, people are discovering the superpower of creating, and a lot of them create just to create, because it is a real and new sense of empowerment. I was not immune. I spun up my fair share of tools I thought I needed, then had to dig through my own bag of slop and throw out about 99% of it. That 99% were just bad ideas. I was reinventing so many wheels that I didn’t have a vehicle that could use them all and still be drivable.

The data backs up the feeling. In Stack Overflow’s 2025 survey, 84% of developers now use or plan to use AI tools, yet their single most common complaint about those tools is that debugging AI-generated code is more time-consuming, cited by about 45% of them. Everyone is generating. Far fewer are shipping something that holds together.

So I toned it down. I went back into the bag of tricks and started putting things on the chopping block. In the end I kept two: a single CSS framework, and a defined list of skills with one agents.md file. Being in a constant state of creation is a drug, and the dopamine fades fast. Most of the tools I was trying to force into one pile needed duct tape and spit to stay together, and that was quietly eating the development time I actually cared about.

Value in What Cannot Be Copied

This points at something that isn’t really new. Back in 2008, Kevin Kelly wrote that “when copies are free, you need to sell things which can not be copied.” AI made the copying free. On Google’s October 2024 earnings call, Sundar Pichai said more than a quarter of all new code at the company was already being generated by AI.

When building gets that cheap, the building stops being the valuable part. The value moves to the things that can’t be generated: knowing what is worth building, positioning it, selling it, and gluing the pieces into a product someone will actually pay for.

This is where I want to talk to the developers directly, because this is the part I think we get wrong about ourselves. The hard part of this work was never the coding. It was the economy around the code. And most of us were already doing that work. When developers sat in the back, we were still doing an enormous amount to bring a product to market. We just didn’t think much about it, because there were teams around us absorbing it.

The same goes for IT people, who technically fix computers but also sit front-facing with clients, make the calls, write the emails, and move information from point A to point B. In essence, we were managing the flow of the whole process. It just wasn’t as visible, or as critical, as it is about to become. If management never made that part of your work feel valuable, that was their miss, not a measure of what you are capable of.

Moving Beyond the Code

So if you have ever touched the work beyond the code, the operations, the client conversations, the selling, now is the time to flex those muscles. Here is where I would start: stop adding to the thing you are building, and go ask the person it is for what they actually need. That one conversation is worth more than the next feature.

The future this points to is one where a single capable person can carry close to an entire company, and what sets that person apart will not be how much they can generate. It will be the quality of the work, how well they understand their clients, and how intimately they can work inside their clients’ products. That part was always ours. AI didn’t take it. It just made it the whole game.

---

References

• Stack Overflow 2025 Developer Survey: AI (84% use or plan to use AI tools; ~45% cite “debugging AI-generated code is more time-consuming” as a top frustration)
• Fortune: Over 25% of Google’s code is written by AI, Sundar Pichai says (Google Q3 2024 earnings call, Oct 29, 2024)
• Kevin Kelly, “Better Than Free” (The Technium, 2008) (“when copies are free, you need to sell things which can not be copied”)