The LLM revolution is a Computer Science revolution
Many companies, driven by downward pressure both from the markets and their own executive teams, are exploring how to integrate LLMs throughout their organizations. Businesses and governments large and small are using large language models for file retrieval, data aggregation, customer support, and even to develop in-house software. Although there are many consequences of this shift (such as AI-generated code quality, the need for human debuggers, skill atrophy, etc.), the focus of this note is how the LLM revolution is shining a light on the overlooked, ubiquitous value of computer science and skilled programmers.
Organizations are converging on LLM use cases that look very different from those the models were initially designed. Big tech is pushing toward models that excel at math, coding, and logical reasoning, whereas academics are probing the limits of current models and questions of agency. The wider world continues to explore what their reducible inefficiencies actually are and are converging on automation and data analysis—both classic computer science and machine learning tasks.
Consider the pattern. Many organizations use LLMs simply to find the right form or document, because their file systems are poorly organized and littered with conflicting legacy versions. Others use LLMs to merge disparate data sources in incompatible formats or to answer statistical questions like "what is the relationship between number of contracts and number of bylaw violations?" These are very different from use cases offered by the Google and Microsoft office suites—text drafting, meeting summaries, basic Q&A.
An important realization here is that these problems are classic computer science tasks. File organization and retrieval, data aggregation, data science tasks (regression, correlation analysis, other forms of machine learning), and automation of all kinds is the bread and butter of CS education. There are well-understood solutions to these problems that do not require LLMs, expensive subscriptions, nor cloud-scale computation. For computer scientists, these are simple tasks that have decades old solutions.
In many ways this “off label” LLM usage is great because it represents organic in-situ identification of the actual challenges that businesses and organizations currently face. LLMs represent a totally new way of interacting with computers. Instead of taking a four-year computer science degree to learn how to make computers do what you want (i.e. learn to code), LLMs allow non-technical folks to construct their own solutions. A more pejorative view for the agentic revolution is that LLMs have finally forced decision makers to realize the benefits of automation and machine learning and that they are worth paying for!
What are the implications? We know models hallucinate. It is becoming clear that AI-generated code is difficult to verify and maintain. And it is risky to build bespoke solutions that nobody in your organization understands. The principal danger is grounded in applying an in-house AI solution to a problem for which it was never intended —a risk especially likely when the designer has no knowledge of how things work under the hood. Will anyone notice the answer is wrong? Or will they just ask the latest model what the problem is? There are parallels here to the large, industrial software that organizations around the world are currently trying to overcome with agentic engineering. Organizations were often forced into expensive, long-term vendor subscriptions for critical software systems that are complex, rigid, and brittle. Will this pattern repeat?
The outcome of the LLM revolution is likely to be a renewed appreciation for computer scientists in every medium and large organization. The rise of unverified AI code and the resulting technical debt create a clear economic imperative: the cost of unmanaged automation now exceeds the cost of hiring people trained to build and maintain it. The next step is hiring computer scientists to solve computer science problems in house (with the help of LLMs, of course). Recent layoffs in the sector notwithstanding, the pendulum will swing back to in-house human expertise building automation tools.
Ultimately, this is less about the tools than about the realization that sustainable automation requires the very discipline organizations tried to bypass. Computer scientists are not just an expense—they are the essential architects of this new operational reality.
In-house expertise is a core value at RLCore. The value of true expertise lies in the ability to develop and deploy right-sized, sustainable solutions rather than forcing an organization to adopt expensive tools and unmanageable automation. Fit-for-purpose tools ensure newly adopted technology serves the organization's goals and that it does not degrade into an operational burden.