原文 · 未翻译
AI means more people can create code. AI also means fewer people will be able to write code.
There are three main incentives to use LLMs heavily. First, the social pressure from the fever tweets about hyperspeed shipping. As over the top as they are, it’s undeniable how quickly working prototypes can be created. If you choose to ignore the tools you will be left behind by the agentic engineers in productivity terms.
Once you try the models, you realise how good they are, and there is the second incentive. These things write working code and as the models get better and better, the argument that they make mistakes will get quieter and quieter until it fades away, like all high conviction opinions that turn out to be wrong over time do.
The third incentive is the strongest of all: human laziness. AI can write the code faster and you can relax with busywork. The model providers are like a gym that makes you Superman once you step inside. Why go to the gym next door where it hurts and you lift measly little weights?
These three reasons mean more code is produced and less is written by hand. And that means the clock has already started on engineering skills atrophying.
And that is an opportunity for the disciplined. The pool of people who would deliberately practice their engineering skills was small before LLMs. The majority of us would only practice when actively interviewing, but writing code in our jobs meant we became more skilled over time. Now that we only ask for the work to be done, remaining skilled has become a very high effort activity. It follows that the number of skilled engineers will drop from what it is today.
And that skill does matter. For the next 10 years, the best companies will still hire engineers who can solve algorithmic puzzles, build systems by hand under pressure, and discuss architecture trade-offs. I believe the best products and breakthroughs will still come from people who deeply understand fundamentals. Super intelligence would break that, but that’s a different essay.
Let’s say you disagree with my idea to practice, and you go all in on agentic engineering as an individual, how might that turn out?
Case one: you are employed, you only write code through agents, your skills atrophy and you move down to a skill level where the pool of people capable of doing what you do has increased enormously. As a software engineer who has forgotten the fundamentals, you now compete with product designers, business people, and your old manager who is “back in the arena”. You are no longer a special wizard of the keyboard, you are just a prompt monkey.
Now the case where you’re an agentic engineer building side products or your startup. If you hit on a product that grows and has a business model, you can raise money, move into management, and build wealth. Great result, no complaints. However, for those that don’t grow, you have created a failed product. The product does not provide the signal that you can build like it did two years ago. It shows very little other than that you prompted an idea into existence and tried to market it. The fever-tweeting VCs of course love this model as they pick up the winners, but its not ideal for the individual.