It's the end of coding as we know it
During a fireside chat at LlamaCon 2025, Meta CEO, Mark Zuckerberg asked Microsoft chief, Satya Nadella about the use of AI for coding. Nadella responded that up to 30% of the code in Microsoft’s own GitHub repositories is written by an AI agent.
Since Microsoft, which is now over 50 years old, began as a company selling tools for software developers, Nadella sees the company pivoting to a place where it is developing tools that can be used by AI coding agents.
Zuckerberg’s vision is a far more aggressive use of AI coding, to accelerate the company’s strategy, based on using AI in a way that keeps people who use the social media’s platforms to remain engaged with content. In doing so, the AI that Meta develops directly contributes to the company’s revenue, by putting more eyeballs in front of relevant advertising.
During the earnings call for Meta’s first quarter of 2025 financial results, Zuckerberg claimed that the ability of AI to write code was reaching the level of a mid-level engineer. By next year, Zuckerberg expects that the majority of the company’s research with AI will be achieved using AI agents.
A quicker return on investment
No one is developing AI systems for fun. Zuckerberg believes that within a few years every company will have an AI agent in customer service, just as they have a company email address and social media accounts.
If, like Meta, these AI systems are being developed and trained to deliver key performance indicators like driving revenue or increasing customer satisfaction, the quicker they can be developed and deployed, the faster the return on investment.
Consider the scenario where a business head wants a new AI system. The choice is use in-house expertise, perhaps complimented by external consulting or submit the task to a large language model that has been optimised to write code.
While some managers may see the benefit in giving the project to the internal team – perhaps due to the initial cost of the AI coder, over time, that cost will come down and human programmers may end up losing out to AI agents that can code quicker and who do not suffer the same kind of hang-ups of real people. While, like human coders, they do make mistakes, they do not have “off days”, need coffee breaks and the social niceties that help them do their jobs better.
What is worrying about Zuckerberg’s remarks is that his view of coding completely eradicates the need for a pipeline of talent to up-skill software developers. There is a case for training human programmers to become more like project managers. But what’s worrying is that the pace of AI progress may even reduce this role’s significance.