Execution Is Cheap. Judgement Is Not.

The ability to write code and create software is being commoditized.

More people than ever can build software with AI. The ability to merely write code is not as valuable as it used to be.

AI agents can write more code faster than you and, with the right workflow and gates, can often do it with surprisingly high quality. They can find and fix bugs before you have even found the right file to open.

This radically changes which skills are valuable.

A year ago, writing code by hand was a larger part of the job. Today, more of that value is moving into directing, reviewing, and understanding the work.

As a software engineer, I still look at code, but I almost never write it myself anymore.

And why would I? By the time we get to implementation, much of the highest-leverage thinking has already happened.

In today's world, with spec-driven development, many technical problems are discovered and solved early in the process. With a human in the loop, assumptions are challenged, constraints are surfaced, and the path is set before the first implementation is generated.

Yes, we can still adjust and adapt. But many difficult problems are already solved by the time we get to coding.

Even when coding is needed during discovery, AI can spin up a few POCs for you to evaluate right then. You can explore multiple options and make a decision early in the process.

So why would I spend my time typing out code when AI can help me get to the important decisions faster?

The Value Is Moving

In light of all that, with old skills going away, what skills are here to stay? What can't AI do? What value do humans still bring to knowledge work?

I personally think there's still a lot of value and it was underlying all the work we were doing before.

I think the biggest sources of value humans bring are judgement and critical thinking skills. When building is cheap, knowing what to build becomes more important. When implementing solutions is easy, identifying which solutions effectively solve the problem and reduce negative side effects becomes paramount. AI can handle the grunt work. We handle the high-level thinking.

But this isn’t that much different than before.

On Judgement

In enterprise software, teams of engineers could solve any problems you threw at them. Many companies floundered due to lack of focus, or, in other words, not knowing what problems to solve. That’s where judgement comes in.

Having a clear vision and good product managers that stick to that vision has always been a key to success in software. Without that, we aimlessly chase the next shiny thing and struggle to provide customer value.

The AI doesn't know what your customers want. AI doesn't know if prioritizing X over Y will bring the most value. It can certainly help analyze the data and inform a decision, but it cannot make that call. One day, maybe it will, but for most of us, that day is not today.

AI doesn't know your customers like you do. It doesn't know your team like you do. It doesn't know your industry like you do. It doesn't know your codebase like you do. Yes, it can ingest any information and context you give it about all those things and leverage that to help you make better decisions, but the decision–the judgement–ultimately rests with you.

On Critical Thinking

A similar thing happens on the technical side.

There are countless ways to implement any given feature or build any given software application. Many teams have fallen prey to endless rewrites because they are chasing the latest, greatest technology and not recognizing what’s good enough to meet the needs of their customers.

Some embark on large refactors choosing to ignore the many obvious downsides of the path they’ve chosen simply because they want to use X framework over another or they figure they've spent enough on this project so they may as well continue.

This is where critical thinking skills come into play. We can ask AI to fix a problem or implement a feature, but we need to recognize what actually fixes the issue vs what is noise and we need to be aware of the downsides. As Andrej Karpathy put it, “you can outsource your thinking, but you cannot outsource your understanding.” Leverage AI to speed things up, evaluate options, and implement solutions, but you need to still understand the problem so you can understand which solution is the one to move forward with.

Getting Aligned With AI

This is why I like the /grill-me skill or /grill-me-with-docs skill so much. By getting to a shared understanding upfront, we can surface all these gotchas and all these assumptions early in the process so we begin evaluating solutions from a shared understanding. AI is great at generating solutions. The trick is getting on the same page so you can actually get the solutions you want instead of solutions that look or feel close but completely miss the mark.

I’ve seen this with many projects I’ve worked on. If I don’t have a great feel for what I want going into it, I often get frustrated that what the AI gives me is not quite what I want.

Often, the trouble there is surfacing what I actually want. I may have an idea in my head but I may struggle to know how to describe it. One real superpower in the age of AI is to clearly, concisely, and accurately describe an idea. The better you can do that, the better your results will be.

More Ideas Will Ship

I saw a picture the other day that I think describes some of these tensions well.

If the meme were honest, there’d still probably be a ton of ideas before AI, but the problem was execution. Execution was a premium service, not a commodity.

Now that execution is a commodity, many more ideas can see the light of day. The question is, which ones actually matter? Which ideas actually solve problems? Which ideas actually impact people and improve life?

If you find one of those, you’ll get the usage you’re looking for.

Conclusion

AI makes coding cheap and easy. AI cannot yet replace judgement and critical thinking skills. It can certainly enhance and augment those areas so they’re better informed and faster, but it cannot replace you.

So keep thinking critically. Keep pushing for understanding. Push your AI to give you better results. Don’t accept what it gives you until you know it solves the problem you want it to solve in the way that will be most valuable and limit tradeoffs.

Execution is only cheap when it is pointed at the right problem. If we use AI to solve the wrong problems faster, we will not get leverage. We will just get more churn.