Is Big Tech pulling up the draw bridge and breaking history? (Part 2)

Computer Programming

Encroachment into and automation of the discipline of computer programming, itself, was not unknown and generally seemed to be in the form of assisting programmers but the palpable blow to programmers when the CEO of Anthropic pronounced in February and then March 2025 (Council on Foreign Relations) the end of their profession was in sight could be felt even by those not directly involved. The opinion was that within 12 months from March 2025 the “AI”-labelled coding programs could largely be left to do the programming with the “small pieces” (the CEO’s description) of higher-level work left to humans until they weren’t needed. In the end, he added, everything could eventually done by “AI” but he couched this outcome as a lesser possibility to a mixture of the framing suggested by the CEO host occurring: that some people would be “able to do some really cool things” and other’s would “be sitting on their sofa collecting UBI [Universal basic income].”

The prediction about programmers is perhaps a slightly revised prediction from that of February when he reportedly said most “software engineering” work could be done by “AI” within six to twelve months. The CEO of NVIDIA said students should consider other careers including education and farming. This same article reported OpenAI’s CEO could see software engineering disappearing at a “gradual but accelerating manner.” CEO’s of Meta and Amazon broadly agreed that computer programming done by humans was on the way out. While the CPO’s of Alphabet and Microsoft argued coding jobs would change as distinct from disappearing according to reports. The Anthropic CEO reportedly identified 50% of “entry-level” programming jobs evaporating by 2030.

These contentions had the appearance of weight because from 2021 a number of increasingly sophisticated “AI” coding programs were released. The following brief account of coding programs concentrates on Big Tech and ignores other developers. Microsoft released the first one as an open source program able to produce both code and documentation. OpenAI followed with Codex, a version of ChatGPT 3, that was able to convert user text requests and turn it into code. This was, perhaps, the first tangible sign of replacement. It was based on code scrapped from GitHub, a private company that was a “hub” where open source code could be stored, shared and edited. Microsoft has owned it since 2018. In June 2021, GitHub Copilot was released as a feature of GitHub and was built by the collaboration of Microsoft, “Github” and OpenAI; Github maintained an illusion of independence from Microsoft. Copilot used Codex as its “AI”. OpenAI discontinued its own Codex. In February 2025, Anthropic announced the availability of a coding tool as a limited release with Claude Sonnet 3.7. Coding had been a feature of some earlier models of Claude but this was a more substantial offering. The tool became fully available in May 2025 with the release of Sonnet 4. It was added as another tool that can be used with Github Copilot. OpenAI re-released an open source version of Codex in April 2025.

Since the Anthropic CEO announcements there has been headline after headline with an Anthropic engineer, its CFO or the CEO saying the “AI”-labelled Claude model writes 80%, 90%, 100% of the company’s code. We don’t actually know what the percentages really mean. There has also been some fallback by Big Tech, including Anthropic, from the extreme announcements as they seemed to realise that most people weren’t actually excited about the prospect of not having a job. Other Big Tech companies weren’t saying “AI” is doing most of their coding.

The upshot is that all this talk and it appears to be largely marketing hype, about “AI” models taking over jobs may be at least, one factor in the reduction of people choosing undergraduate degrees related to computer coding, computer programming or software engineer in Australia, the UK, Canada and the USA. Offers made to study Information, Communications and Technology (computer science related courses including “AI”) in Australia in 2025 were down by about 9% and the trend likely continued in 2026 (ABC 29/1/2026, Australian Department of Education). Overall, 3% less individuals were accepted into computer courses in the UK in 2025. Computer Science acceptance fell by 10% and Software Engineering by 5%. In contrast, there was an increase in related computer courses such as Information Technology and Artificial Intelligence (bcs The Chartered Institute for IT Demand for UK computing degrees in 2025 website. The lower numbers were particularly evident with younger students. The number of 18 years olds accepted into computer-related courses fell by 9%.

In the USA, computer science student numbers at fall enrolments, starting in August/September to commence study in 2026, dropped by between about 3.8 to 8% (depending on type of course) and those enrolling as a graduate dropped 14% in 2025 (National Student Clearinghouse Research Center statistics, final fall enrollment trends, January 2026. Spring 2026 enrolments also appeared to continue in this downwards direction. A survey of relevant members of the Computing Research Association found less students enrolling in the fall season in Canada as well as the USA. One of the reasons given was students were concerned about the loss of jobs due to “AI”, particularly in software engineering. Other reasons were: some choosing to learn about about the infrastructure relevant to running software, like electrical engineering; some choosing to study computing in areas relevant to “AI”; and; a drop in overseas students.

Software engineering, software development and coding can be considered synonymous to computer programming but some use the first two terms to refer to higher level abilities such as planning and design. In the USA official labour statistics, software engineering is consider under software development but sometimes software engineers can be in charge of developers. Coding often appears to point to the filleted remains, the coded instructions, once design and planning towards a wholistic end, for example, an application, have been removed. The Anthropic CEO doesn’t appear to have mentioned “software developers” in this context. Perhaps that means nothing. He did, whether he meant it or not, give a breathing space to CEOs who would presumably consider themselves as highly skilled and possibly those whose job couldn’t be digitised because of its physical nature when he effectively said not everyone would lose their job. The CEO’s of industry probably envisaged themselves running their industries just talking to a model which would then go off and build a “cool” app. Perhaps they wouldn’t even need the Anthropic CEO. Maybe the Anthropic CEO left a little bit of wriggle room to deal with this type of situation: wouldn’t the “AI” still need a “developer” who could continue its development into adapting to physical work?

There are reasons to challenge this view of an “AI”-labelled model led future. The Anthropic CEO’s predictions contrast strongly with the reality described by those of the USA Bureau of Labor Statistics for computer programmers and software developers. Computer programmer jobs would still number over 100 000 in 2035 representing a loss of 7% over ten years from 2025. Software developer (combined with testers and quality assurance analysts) jobs that numbered in the near 2 millions in 2025 would grow by 10% over the ten year period. The web pages were dated 27/8/2026.

Research can be found to challenge the views of the Anthropic CEO and their pronouncements should always be critically evaluated to reduce being swayed by marketing spin. There is some research pointing to young workers bearing the brunt of the loss of jobs because of “AI” use by companies particularly in the USA but at least one study also considering Australia, Canada and the UK – perhaps a way of quieting the fears of more experienced people that near everyone was about to loose their jobs. The research is in its early days and should be read with caution. Some questions should be considered when reading research in this area and are derived from points made by a sociological expert and a relevant paper and these are:

• are assumptions sound? For instance, it doesn’t seem likely employers would stop hiring young workers immediately ChatGPT became available or even before it became available.
• are all variables included in the study? For instance, could gigification of work explain the job losses as the jobs are now undertaken by young ‘contractors’ rather than ‘employees’.
• is the data used representative of employment as a whole? For instance, only using one company’s payroll is a limitation when the company only focuses on one type of job not all types of jobs.
• if the data uses “big data” that is estimated from a chatbots “chat” how accurate is it? For instance, it is argued that using words entered into Anthropic’s “AI”-labelled Claude model as definitively related to specific tasks and then grouping those tasks into specific jobs such as is done to produce the “Anthropic Economic Index” is inaccurate.
• has the paper been peer-reviewed?

It seems conclusions from this type of research would be more solidly supported if longer term and more comprehensive analysis based on appropriate data was undertaken. The results of these entry-level decline studies don’t necessarily mesh with what businesses are indicating their uptake of “AI” is and should be considered with the continuing errors (anthropomorphicly known as hallucinations) and the advice by experts of the need for human oversight of “AI”-labelled code. Similar evaluation of Big Tech announcements should be made.

If we follow the Anthropic’s CEO’s line of thinking to its natural conclusion with Big Tech “AI”-labelled products doing all the programming the consequence will be a loss of community knowledge of what programming is, what it can do, how it is done and what has been done before.  Ending up with only “AI”-labelled products seems unlikely though and there still may be programmers, perhaps in other countries, and also hackers. Big Tech “AI”-labelled products aren’t known for their creativity due to the limitations of the companies wanting anthropomorphised machine learning and the data used. Could the remaining human programmers work out a way to counter the Anthropic’s CEO’s, presumably, ideal future? Could hackers be the winners?

If Big Tech gets their way, programming will become increasingly automated and the need for programmers lessened by the encroachment of models able to translate ordinary English into programming code. There will eventually be no need for programmers as all their work will be replaced by the models. Computer programming will become dominated by the Big Tech companies, with fewer people choosing to learn computer programming. The history of computer programming that programmers hold in their heads, their methods and their documents, will recede into the background replaced by the history of Big Tech and its models.

Continued tomorrow… USA Open Source