The New Economics of Generating Code
The next is replace -- replace feature after feature after feature of the older Cerner system with a new Cerner system, new Millennium, which we are not coding in Java like we usually do. The new Cerner system is being generated -- as you know, generative AI generates code. We have an application generator called APEX. And we are not writing code for the new Cerner; we are generating that code in APEX, and it's going extremely well.
This is a quote from Larry Ellison in Oracle’s latest earnings call. It should be taken with a grain of salt — Ellison is a master of narrative, and he’s addressing an audience of investors. Whether APEX works as well as he claims or if developers are simply using GitHub Copilot, the fact remains: this is the future of a good chunk of software development.
The cost of refactoring legacy codebases is dramatically decreasing. The StackOverflow developer survey consistently puts Fortran, COBOL, and Assembly developers in the middle of the pack with regard to salary. In 2022, while the median salary for most languages remained the same, COBOL developers saw the most significant salary jump.
Critical systems still run on mainframes in banks, airlines, and the public sector. It’s easy to see how this is the case: the most promising use cases for software were written many years ago, and the more critical the system, the lower the chance that management (or developers) were willing to mess with a working system.
There’s a short supply of developers willing (or, more importantly, able) to deal with these systems or languages.
LLMs provide an avenue to translate many of these languages to modern systems. It’s not automatic — and might never be, but it can drastically reduce the cost of performing a migration like this.
On a lesser scale, there are many other opportunities where migrating a codebase can be very economically beneficial.
Acquired companies. When a software company is acquired, integrating or migrating the codebase is usually one of the first priorities. Some acquisitions can continue to run autonomously for a while, but most will at least need to be migrated to the company’s infrastructure.
Cloud providers can acquire companies with significant infrastructure spend on another provider and migrate the spend to their own platform.
Private equity investors can modernize the code base enough to either (1) reduce the ongoing maintenance cost or (2) put the company in a better position to ship future products or integrate more cleanly into a more strategic acquirer.
Strategic acquirers can integrate codebases with their internal technology much faster. This means sharing infrastructure, monitoring, code, toolchains, and more.
Existing technical debt just got cheaper. Within companies, the cost of technical debt just dropped substantially. Refactoring with AI assistants is much quicker with Copilot, and we’re on the verge of seeing more automated systems that autonomously fix or maintain software. It’s not here yet, but we can already see the prototypes of it in automated code review bots, dependency management bots (dependabot), and programs that do housekeeping tasks like triaging, labeling, assigning issues, resolving merge conflicts, and more.