🔍 Read the full analysis: A Closer Look At Whether Students Should Learn To Code on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
AI tools now author a large share of code at some companies, but that does not establish that learning programming is unnecessary. Research cited in a 2026 review finds mixed evidence on speed and suggests that using AI while learning can weaken some students’ understanding; the entry-level job market is also tight, though software job postings have recently risen.
A November 2026 review of AI coding adoption, developer experiments and hiring data concludes that students should still learn programming, while placing more emphasis on specifying, reading, testing and debugging code. The evidence does not show that AI has made programming knowledge unnecessary: it shows that tools can produce code, while productivity effects, learning outcomes and entry-level job prospects remain uneven.
Anthropic reported that in May 2026, Claude authored more than 80% of the code the company merged into its codebase, compared with a share in the low single digits before Claude Code launched in February 2025. The company said engineers directed and reviewed the generated code. It also reported that the typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024, but cautioned that lines of code measure quantity rather than quality and that the figure almost certainly overstates the productivity gain.
Independent evidence on speed is less settled. In a randomized METR trial involving 16 experienced open-source developers and 246 tasks in codebases they knew, developers using AI tools available from February to June 2025 took 19% longer to complete the tasks. Participants had expected a 24% time saving and later believed AI had saved them 20%. METR said a later study offered only weak evidence about whether newer tools increased speed: its estimates included the possibility of no effect.
Learning results raise a separate concern. In an Anthropic experiment described in the review, mostly junior developers learning a new library scored 50% on a follow-up quiz when they used AI, compared with 67% among those who coded by hand; the largest gap was in debugging. On hiring, Indeed data showed entry-level roles made up 4.5% of US software-development postings in the first quarter of 2026, while senior roles accounted for 69.3%. But postings were up about 23% year over year in November, according to the source material.
Programming Skills Beyond Code Generation
For students, the distinction is between producing code and understanding it. AI may take over more typing, but people still need to decide what a program should do, how it fits into an existing system, whether its output works and whether its design is secure. Those skills matter when generated code is incomplete, unsuitable or wrong.
The evidence also bears on how schools teach computing. If students rely on AI before they have built a working grasp of programming concepts, they may finish exercises without learning how to diagnose errors or explain their decisions. The reported quiz result is one experiment, not proof that AI always harms learning, but it points to a practical risk teachers and students can test for.
For people entering the workforce, the figures suggest a harder first step than the idea of an expanding technology sector might imply. A small share of postings was entry-level in early 2026, even as overall software-development postings later recovered. Students may need to show they can review and validate AI-generated work, not just prompt a tool or produce a working demonstration.
Products Worth Considering

Amazon Kindle Scribe Colorsoft 64GB with Premium Pen

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue

Samsung 32 Odyssey G5 QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From AI Forecasts to Measured Results
The debate gained attention after Anthropic chief executive Dario Amodei said in March 2025 that AI could write 90% of code within three to six months and essentially all of it within a year. Anthropic’s later internal figures show extensive use of Claude at the company, but they do not establish that the same level of use or productivity applies across the software industry.
Other figures provide a different scale and time frame. Google chief executive Sundar Pichai said in October 2024 that AI generated more than a quarter of new code at Google, which engineers then reviewed and accepted. METR cited an estimate in February 2026 that about 4% of GitHub commits were authored by Claude Code. These measures track different organizations and activities, so they are not direct comparisons.
Experimental results also depend on the work being done. A 2022 test involving 95 professional developers found that participants using GitHub Copilot completed a small JavaScript HTTP-server task 55.8% faster on average. METR’s later trial involved real tasks inside large projects familiar to the developers. The contrast cautions against treating a result from a short, self-contained exercise as a general estimate of speed on everyday software work.
“The programmer still needs to specify what is being built, how it fits with other code and whether this is a secure design or an insecure design.”
— Dario Amodei, Anthropic chief executive, speaking at the Council on Foreign Relations on March 10, 2025
How Broad Are These Results?
The available figures do not settle how much AI improves software productivity across employers. Anthropic’s code-usage statistics come from a company that builds AI coding tools and uses them heavily; its own caveat says code volume is not a direct measure of useful output. The METR trial was small and focused on experienced developers working in projects they already knew, while its later estimates had confidence intervals that included no effect.
It is also unclear how much of the reported decline in employment among younger workers in AI-exposed jobs is caused by AI. The review cites Stanford researchers’ finding that employment for 22- to 25-year-olds in those jobs was 19% below where it otherwise would have been. That estimate does not, by itself, show that AI caused the shortfall. The cited job-posting shares describe a particular platform and period, not every employer or the number of hires made.
The learning experiment involved a specific library and group of mostly junior developers. The material does not establish whether different teaching methods, levels of AI support or longer-term practice would produce the same quiz gap. Nor does it show that students who use AI cannot learn programming. Those limits make the results a warning about how AI is used, rather than a verdict on classroom use as a whole.
What Students and Schools Can Track
Students deciding what to study can treat programming as a skill set that includes more than typing syntax. Practice should include writing code, explaining how it works, testing edge cases, debugging failures and reviewing AI suggestions against the intended requirements. Schools can compare learning with and without AI by checking whether students can later solve related problems and explain their choices independently.
For educators and employers, the next useful evidence will come from larger, repeated studies that test newer tools on varied programming tasks and measure working software, not just lines produced or time spent. Labor-market data over additional quarters can help show whether entry-level posting shares and overall demand continue to change. Until those results are clearer, the available evidence supports learning to code while treating AI as an aid that still requires informed human direction and review.
Key Questions
Should students still learn to code?
Yes, the evidence summarized here supports learning programming, including how to specify tasks, read code, test behavior, debug problems and judge design. AI-generated code still needs human direction and review.
Does AI make software developers faster?
The findings are mixed. In METR’s early-2025 randomized trial, experienced developers took 19% longer with AI on the tested tasks. METR’s later evidence suggested possible improvement with newer tools but was too weak to establish the size of any speed gain.
Can using AI affect how well students learn programming?
One Anthropic experiment found that mostly junior developers who used AI while learning a new library scored 50% on a follow-up quiz, compared with 67% for those who coded by hand. The experiment was limited to a particular learning task and does not establish that AI always reduces learning.
Are entry-level software jobs disappearing?
The cited Indeed data show entry-level roles accounted for 4.5% of US software-development postings in the first quarter of 2026. The source also reports that postings overall were up about 23% year over year in November 2026. These figures do not establish that entry-level jobs are disappearing or that AI caused the posting mix.
Source: ThorstenMeyerAI.com
