One striking feature of modern artificial intelligence, especially in large language models like ChatGPT, Claude, and Gemini, is their ability to instantly generate surprisingly advanced software engineering solutions. One interpretation of this phenomenon is that we are in the midst of a wave of commoditization. A specific subset of cognitive work, once scarce and expensive, has become infinitely accessible at minimal cost. With the rise of “vibe coding”, where AI tools are used transform a basic idea into a functional app or website almost instantly for fun, it seems that any basic concept for a digital product can now be cloned on the spot. The barrier to entry has dropped dramatically, while the bar for uniqueness has risen just as fast. Any simple enough product can be replicated instantly. Are we witnessing the rapid commoditization of cognitive work, with software engineering at the forefront?
{: .float-left }
It sounds eerily familiar. It sounds like the manufacturing industry’s conundrum over the past fifty years. For every successful product, there are fifty knockoffs. Creating something truly unique now demands far greater resources. And to stay competitive against imitators relying on AI, software entrepreneurs will be compelled to use generative AI themselves. There is no way around it. Development velocity has exploded and it will not go back. It is not that this kind of change has not transpired before. It is merely twenty years since software development relied on printed books and documentation. Who in their right mind would use only that today? At the same time, Google Search and later Stack Exchange bolstered programmer efficiency, not to mention the collaborative tools like, dare it say it, Jira and Git.
There is an old business saying: execution is everything. An idea is a dime a dozen and does grow on trees, but making it happen and turning it into a business is something else. But with artificial intelligence, does this still hold true? Or has execution suddenly become… nothing? The legendary Intel CEO Andy Grove has said it, Amazon founder Jeff Bezos has said it, and countless other leaders: “Execution is everything”. Often preceded by “Ideas are nothing”. Execution refers to the effective implementation of strategies and ideas. It is the “How?” to the idea’s “What?”. A million-euro idea is still worth nothing unless it is made tangible. The part between the idea and the result is the execution. It is often a messy combination of people, roles, organization, conference rooms, buildings, and customers. Not to speak of the finance, marketing, compliance and what not that is essential to getting a product out the door. The overhead, so to speak. Any technology sufficiently able to reduce this overhead will be adopted given time.
So, where does this take us? Will 90 percent of software be developed by AI already during 2025, as Dario Amodei, CEO of Anthropic, has said? His assertion is likely true, but likely not for the reasons one initially would assume. I would even go so far as to say that AI already produces 90 percent of all code produced, but that a large chunk of that is not new, unique code, but single-use code, repeating existing code and multiple iterations while clarifying prompts. On the current trajectory, a large percentage of code will soon have been AI-generated at some point. The software engineer going from executee to executor, from being a programming specialist to a programming generalist with domain-specific knowledge. From being a coder to being a business person orchestrating the execution of coding workflows.
Software engineering may just be one of many such cognitive business spaces being commoditized. It is mainly due to the availability of abundant data on software engineering tasks that it is the first to go. When companies realize that any general cognitive work is much the same, and gather the data, the same thing will unfold in other lines of cognitive work.
There is a larger question here. What will happen with the people?
Andreas Påhlsson-Notini, a@nial.se