You’ve no doubt read dozens of posts about AI taking or not taking your software developer job. But you haven’t read one written by me, so here we go. Now I know what you’re thinking. Gee Mike, you’re as old as dirt. What the hell do you care whether AI takes my job? You’re going to retire your ass in a couple of years anyway. Well, you’re right. Suck it, losers! Wait, did I write that out loud? I didn’t mean to.
Is AI (and in this context, AI refers to large language models (LLMs)) capable of taking your job? It is according to the breathless rhetoric we’re hearing from tech bros. But let’s get some perspective. Every technology has a hype curve known as the Gartner Hype Cycle, and large language models are no exception. Here’s what the hype curve looks like.
We can divide this curve into three sections (with apologies for the technical jargon I’m about to throw at you):
- The Holy Shit! phase. This is the part from the origin to the Peak of Inflated Expectations. In this phase, LLMs can do everything. They can write your code, they can write all of your tests; there are no limits.
- The Aw Crap! phase, from the Peak of Inflated Expectations to the Trough of Disillusionment. This is where we realize that LLMs can’t do everything. They are not going to generate all of our software applications for us. The hype has evaporated. There are limits to what the technology can do.
- The Yeah But phase, as in “yeah, but the tech is still pretty useful though.” This is the Slope of Enlightenment and Plateau of Productivity parts of the curve. At this point, we have figured out what LLMs can really do for us and how they will impact our jobs. There are commonly-accepted standards and best practices for using LLMs as a tool in your daily work.
Where are we on the hype curve? Most people are probably on the upper part of the rise towards the Peak of Inflated Expectations. However, some of us may have already passed the peak and started the decline to the Trough of Disillusionment.
A recent MIT report shows that 95 percent of companies have seen no real return on their LLM investments. A survey of developers who are using LLMs has found that far from the 10x improvement in productivity that some people have predicted, they are actually seeing about a 20% to 30% improvement in productivity on average. And this year’s K-Prize competition winner crafted LLM prompts that were so good that the LLM he used successfully addressed a whopping 7.5% of the GitHub issues in the K-Prize AI coding challenge. Ouch!
Vibe coding is another trend that is being hyped lately. In this approach, LLMs are used to generate entire apps. If people don’t know how to code very well, no problem. LLMs will do it for them. But vibe coding isn’t really a thing. All it has actually accomplished is to create an industry dedicated to fixing the crappy code that LLMs have generated during vibe coding sessions. It turns out that if you don’t know how to write an app, you’re going to struggle to describe an app well enough and in enough detail for an LLM to write it.
Yet CEOs are toying with the idea of replacing their junior software developers with agentic LLMs, while their senior developers will be reviewing the work generated by the new AI developers. After reading the previous paragraphs, you may have arrived at the conclusion that this strategy can’t possibly work. And you would be correct. There are a couple of reasons why this is so.
The entirety of human history has shown us that when a new technology is introduced that can significantly increase production throughput, successful organizations don’t maintain their prior productivity with fewer human workers. They increase their productivity with the same staff. The product or service they create becomes less expensive to produce and thus becomes affordable to far more consumers. They tremendous increase in potential customer base creates far more income. Introducing LLMs to software organizations won’t result in the same amount of software being produced by fewer human developers. It will result in more software being produced by the same staff augmented by LLMs. Companies that don’t realize this will find themselves out of business.
The idea of replacing junior developers with agentic LLMs isn’t limited to the software industry. Some law firms are thinking about replacing junior partners with LLMs. Given the current state of LLMs, this strategy won’t work in the near term but it doesn’t make sense in the long term either. Where do senior developers come from? They come from junior developers. But if junior developers are replaced by agentic LLMs, then either somehow these LLMs graduate to senior developers (then who reviews their code?), or organizations will run out of senior developers and will have to hire them externally. Their salary demands would skyrocket. Want to pay half a million or more per year for a senior developer? And not an elite one, just an average one.
So will AI take your job? There may be some job losses in the short-term as CEOs who have bought into the 10x productivity hype replace people with agentic LLMs. However, when they realize that this strategy won’t work, they will either hire people back or it will be too late and their company will have gone under, having been outcompeted by rivals who understand what LLMs can really do. In the long run, AI will not take your job, but it will change your job, as new productivity tools always do. If you are willing to learn and use LLMs effectively to increase your productivity, you’ll be fine.