History On Vibes And The Future Of Coding

Victor Erukhimov, CEO of CraftStory.

getty

I like to read about history in books rather than to watch it with my own eyes. However, time and again, I can’t help feeling that I live during historic changes. The new economic disruptor seems to be vibe coding. It is expected to have a very significant influence on the economy. A term that sounds like a chill weekend project should not be used to brand a technological change that is compared to the Industrial Revolution, but here we are. Thanks to Andrej Karpathy’s tweet, vibe coding now means, among other things, a real threat to software developers’ job security.

Mark Zuckerberg, the CEO of Meta, said, “Probably in 2025 we at Meta are going to have an AI that can effectively be sort of a mid level engineer…that can write code.” Marc Benioff said in December 2024 that Salesforce is “not adding any software engineers next year.” The just-released AI 2027 document predicts a complete stop on hiring software engineers in 2027. The landscape of AI coding tools is already vast: Cursor, ChatGPT, Replit, Zencoder, Lovable and many others. There are between 26 and 29 million software developers in the world, according to different estimates, and if most of them lose their jobs, this will have a significant impact on the economies of some countries. However, how soon will this happen, and how bad is it going to be? Let us try to answer this question.

Will It Work?

Obviously, the adoption of vibe coding will depend on the quality of code generated by AI, it is still around the level of a junior/mid software engineer. However, the progress in this area is so fast that it is reasonable to assume we will soon have AI that codes like a senior software engineer, capable of correctly interpreting instructions in natural language, making few bugs and writing tests to find them out and fix them. Even between the time I write this paper and the time it will get published, there will definitely be significant improvements in the AI code generation. So, let’s instead focus on another limiting factor: the responsibility for the code.

My previous company was developing automotive safety systems from 2012 to 2016 (and was acquired by Intel in 2016). When we just started, I heard countless predictions of self-driving cars arriving in 5 years and all taxi drivers and truck drivers losing their jobs. We do have self-driving cars now but in very limited settings. One of the leaders in this space, Waymo, operates only in metro areas of a few US cities. Why is that? According to a paper published in Nature Communications, self-driving cars are safer than human drivers, so why are we still driving our cars? One of the top limiting factors is regulation uncertainty and liability: In the event of a car killing somebody, who is going to be responsible? Right now, it is the driver operating the car. Will the manufacturer be responsible for all deaths caused by self-driving cars, however few there are? Another limiting factor is the public opinion: We, the drivers, don’t like to wait behind a self-driving car that is too polite to squeeze into dense traffic.

A similar argument can be made for an AI software developer: If an AI writes code, who is responsible for bugs in this code? What if this is bank software that gets hacked, and people lose money? What if this is automotive software or plane software, and people get killed? According to this logic, the first adopters for the AI coding will be products without significant potential liability issues, such as prototypes, landing pages, support, computer games. Adoption in software that designs buildings and flies planes will be much slower.

Obviously, it’s not black and white: In high-responsibility areas, AI will probably still be used as a coding assistant and will still increase productivity, or AI will be used to write code while humans will write tests. In addition, a customer usually won’t know how much of the code was written using AI, so there won’t be any public outrage, as is the case with self-driving cars.

How Companies Should Respond

So, given all of the above, can we make a prediction on how soon the software industry will adopt the coding tools, assuming that they work perfectly? My gut tells me that nimble companies will be quick to adopt the new tools, as they will give them a competitive edge. Any repetitive software development job that does not bear a significant degree of responsibility, such as safety or compliance, is going to be at risk in the next couple of years. Other companies will be slower to adopt but will eventually cave to the competitive pressure. It seems that there will be a significant number of software developers who will lose their jobs in the next few years. So, what is going to happen to them and the industry as a whole?

As with other technological revolutions, it is highly improbable that the software market won’t grow rapidly because of this change. We can fully expect the Jevons Paradox to work for the software development market. Many applications can be enabled by the cheaper software development costs, including better healthcare IT systems that are patient-friendly for a change, more hardware devices (because writing software for them will be cheap), investment apps, an abundance of software, modernized government IT systems, better banking/educational software and so much more! Creating these products will require a lot of jobs, filling many, many gaps that vibe coding won’t be able to.

Of course, this change will be accompanied by turbulence. Companies will die, and new ones will arise. History shows time and again that in times like these, countries that embrace a tech revolution and allow creative destruction to run its course win over societies that try to protect themselves from the change. Let’s make some history on vibes.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?