Nails used to be made by hand. Then machines came along and nail makers became machine operators. If any of this sounds familiar, it should.
We are still better at reasoning than agents are. Humans have a much larger context window than AI, and I don't mean tokens. I mean the ability to hold three different domains in your head at once, cross-reference them, context-switch, and produce something that actually sits at the intersection. Agents are remarkably good at single-domain work, but the moment you need to pull threads from economics, engineering, and product strategy into the same argument, they fall apart.
Agents Accelerate ExecutionLink to this section
Where agents genuinely excel is execution. Use them to help with research, guide them through the domains and context you've already mapped out, and let them do what they're best at: multi-tasking implementation at speed.
This means that if you enjoy writing code itself but find the problem solving boring, AI agents won't do much for you. You'll keep reviewing their output and finding wrong code, dumb approaches, nonsensical decisions. And the issue is: that is your fault. You did not create a good plan beforehand and everything went haywire.
Personally I've never liked coding all that much and, in fact, I was never the best of programmers. But, and this is an important but, I've always been a really good problem solver. Most of my time working in Machine Learning Engineering was invested in figuring out how to solve problems the company had in a way that left the system better than I found it. This might sound obvious, but we all know it isn't. We all know that person who gets stuck on the most pedantic things possible.
I never focused all that much in improving this skill as well because, to me, programming is the easy part. Planning is where the high wall is. The actual implementation is more of a mechanical problem than anything else, and AI agents, while not good at planning, are the best mechanics I've ever seen. They can type as fast as ten people with reasonably OK code quality.
Of course, since they are not good at problem solving, if the person responsible for planning their work did not do a good job, they'll fail at the typing task too.
But even when the plan is good and the agents deliver, I do understand that some might like to write code themselves. Just to get this out of the way, I don't have a problem with that and do see the reason behind it. Just like how I like handcrafted furniture, or how I like to do the drawing myself rather than through an AI, some people actually enjoy the craft of coding. Well, the point I am trying to make is that I think coding yourself will start to become an artisanal task, a hobby if you will.
That said, just like good carpenters are still valued even though mass produced furniture exists, I do think really good software developers (much better than me) will still code by hand from time to time. At least for a time, while AI still hasn't quiet got to their level. Since I am not one of those programmers, I've been an avid adopter of AI since around 2022! At the start just for autocomplete, but as soon as more complex things became possible I kept going deeper. I feel like I have never been this productive in my life and I don't think I've coded anything since at least April last year. Maybe before.
And exactly because of this huge buy-in on AI agents, I keep finding myself thinking that, cliche as it sounds, software development is going through something very close to an industrial revolution.
The Nail IndustryLink to this section
For a long time software development was a craft industry: labor intensive, hard to scale with capital, dependent on scarce talent, with demand consistently above the supply of capable hands. That scarcity kept wages high and made the whole sector structurally artisanal. Then a new production technology arrived and like every early mechanization wave, it first looked worse than the craft it threatened. The best artisans still produced better results.
Some old timer said that was proof that the machanization amounted to nothing. Other saw the value in it and pivoted.
If you don't think that is true, take the example of the nail industry: nails were once made by hand, which meant production was expensive, slow, and constrained by skilled labor. Supply was naturally limited, quality varied with the worker, and demand remained bounded not only by use but by cost.
Then mechanization entered the process!
At first, machine-made nails were not as good as the artisanal product, and the best craftsmen still clearly outperformed the new method. The revolution started NOT at the moment at which the machine became better than the best artisan, but when it started to be good enough, cheap enough, and scalable enough to make the difference between the average laborer and the best artisan blurry. As each marginal improvement in throughput, consistency, and quality compounds on the last, costs fall, output rises, and demand expands because what was once too expensive to use broadly becomes cheap enough to use everywhere.
The old artisanal method does not vanish but it loses its position as the center of the industry. What replaces it is a different labor structure: less hand production, more machine operation, more process control, more engineering around the production system itself.
Well, hello fellow nail maker.
Don't Outsource the ThinkingLink to this section
AI is the machine. It will come for our profession, and that's okay. The issue is that we need to reinvent ourselves and understand how to stay on the right side of this trend.
We are now in the middle of the S-curve and we are already seeing the results of that mechanization wave. I am sure things will only grow from here and not many will be capable of saying they are better than the machines that replaced then. But that is almost beside the point. Industrial transitions do not begin when the machine surpasses the best worker: they begin when the machine becomes good enough, cheap enough, and fast enough to change the economics of average production.
The typer will be replaced by the PM who can now ship, the analyst who can now build their own tools, the middle manager who no longer needs to file a ticket and wait. That part is already happening. The engineers and researchers, though, will stay, and their job will be what it always should have been: reasoning, solving problems, and directing the work while agents implement their solutions. The value moves from the hands to the head.
Don't outsource the thinking.