🔍 Read the full analysis: What AI Changes—and Doesn’t—About Learning To Code on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
AI is writing a large share of code at some companies, but available evidence does not show that coding knowledge has become unnecessary. Studies report mixed productivity results, fewer entry-level software postings and weaker learning outcomes in one AI-assisted experiment; demand for software developers has also recently risen.
AI is generating a growing share of software code, but evidence available in 2026 does not show that people can skip learning to program. Data and studies cited in a November 18 report by Thorsten Meyer point to a more complicated picture: AI can produce code, yet measured speed gains vary, entry-level hiring is weak, and one experiment found that developers learned less when they relied on AI.
Anthropic reported that more than 80% of the code merged into its codebase in May 2026 was authored by Claude, compared with what the company described as the low single digits before its Claude Code tool launched in February 2025. The company also said its typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024. Anthropic cautioned that lines of code measure quantity rather than quality and that the eightfold figure is “almost certainly an overstatement” of productivity gains.
Evidence on whether AI makes developers faster is mixed. In a 2025 randomized trial, the nonprofit METR studied 16 experienced developers completing 246 tasks in codebases they already knew. Developers took 19% longer when allowed to use AI tools, even though they expected the tools to help and later believed they had made them faster. METR said its follow-up evidence, gathered as tools improved, was too weak to establish the size of any later speed gain.
Hiring and learning data raise separate concerns. Indeed figures cited in the report put entry-level roles at 4.5% of U.S. software-development postings in the first quarter of 2026, while senior roles made up 69.3%. However, postings were up about 23% year over year in November, after reaching a low in May 2025. In an Anthropic experiment, mostly junior developers using AI to learn a new library scored 50% on a follow-up quiz, compared with 67% among those who coded by hand; the largest difference was in debugging.
The Skills Behind AI-Written Code
The figures suggest a shift in what coding work involves, not proof that programming knowledge is obsolete. If developers increasingly direct AI-generated code, they still need to specify the task, understand how a change fits an existing system, test the result and identify defects. Those abilities depend on knowing enough about software to evaluate what the tool produces.
For students and career changers, the hiring mix makes the route into the profession less straightforward. A smaller share of postings is explicitly entry-level, while the available data do not establish why that share has fallen or how much AI is responsible. Employers and training programs face a related question: if junior staff get fewer chances to write code themselves, they may need deliberate practice in debugging, code review and system design, rather than relying on generated output as a substitute for learning.
The reported learning experiment is not a verdict on every AI tutor or coding task. It does show that finishing work with an assistant and understanding the material are different outcomes. Readers should treat productivity claims and learning results as dependent on the task, user experience and how the tools are used.
Products Worth Considering

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

Samsung 27-Inch Odyssey G5 QHD Gaming Monitor (G50F)

Logitech MX Creative Console with 9 Customizable Keys, Graphite
As an affiliate, we earn on qualifying purchases.
From Code Generation to Review
In March 2025, Anthropic chief executive Dario Amodei predicted that AI would write 90% of code within three to six months and essentially all of it within a year. Those time estimates were predictions, not measured results. In the same remarks, he said programmers would still need to specify what was being built, how it fit with other code and whether the design was secure.
Other company figures offer a different scale. Google chief executive Sundar Pichai said in October 2024 that AI generated more than a quarter of new Google code, which engineers then reviewed and accepted. METR cited an estimate in February 2026 that Claude Code authored about 4% of GitHub commits. These measures cover different organizations, time periods and definitions, so they should not be treated as directly comparable industry-wide shares.
Experimental results also depend on the work being tested. A 2022 study of 95 professional developers found that participants using GitHub Copilot completed a small JavaScript HTTP-server task 55.8% faster on average. METR’s later study involved real tasks in large codebases familiar to experienced participants. The contrast illustrates why a result from a short, defined coding exercise cannot automatically predict performance on maintenance work in an established project.
“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 in March 2025
What the Evidence Cannot Yet Show
The available figures do not establish how much AI changes productivity across the software industry. Anthropic’s internal data come from a company that builds AI tools and uses them heavily; its coding-share and output measures may not represent other employers. METR’s 2025 trial covered a small group of experienced open-source developers and particular tasks, while its later study produced estimates with confidence intervals that included no effect.
It is also unclear how much AI explains the low share of entry-level postings. The report cites Stanford researchers’ finding that employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it otherwise would have been, but that finding alone does not establish that AI caused the difference or that it applies specifically to software hiring. The cited figures do not show whether recent growth in software postings will create more opportunities for beginners.
The learning experiment involved a particular new-library task and mostly junior developers. It does not settle whether structured AI use, such as asking for explanations and then testing understanding, improves or harms learning over time. Longer-term outcomes for students and workers remain uncertain.
Hiring and Learning in the Next Data
The next useful evidence will show whether software job postings continue to rise and whether entry-level roles recover alongside overall demand. Future employment data may help distinguish a short-term hiring cycle from a more lasting change in how companies recruit and train junior developers.
Further controlled studies can test newer AI tools across different experience levels, tasks and codebases, while measuring both completion time and the quality of the resulting work. For education, follow-up research can compare ways of using AI with hand-coding and assess what learners retain, especially in debugging and code review.
Until those results are clearer, the evidence supports a measured conclusion: learn programming, but do not treat typing code as the whole skill. The current data put understanding, specifying, testing and reviewing software at the center of adapting to AI-assisted development.
Key Questions
Does AI mean people should stop learning to code?
No. The cited evidence does not show that coding knowledge is unnecessary. It points to continuing value in understanding software, specifying changes, testing generated code and debugging errors.
How much code does AI write at Anthropic?
Anthropic said Claude authored more than 80% of code merged into its codebase in May 2026. The company also warned that its code-volume figures do not directly measure productivity or quality.
Has AI been shown to make developers faster?
Results vary. METR’s 2025 randomized trial found experienced developers took 19% longer on the tested tasks with AI available. A separate 2022 Copilot experiment found faster completion on a small coding exercise. METR said later evidence about newer tools was weak.
Are entry-level software jobs disappearing?
The cited Indeed data show that 4.5% of U.S. software-development postings were entry-level in the first quarter of 2026, but they do not prove that AI caused the low share. Overall postings were up about 23% year over year in November 2026.
Can AI help people learn programming?
It may, but the cited Anthropic experiment found that mostly junior developers using AI to learn a new library scored lower on a follow-up quiz than those who coded by hand. The result covers one experiment and does not determine the effect of every way of using AI for instruction.
Source: ThorstenMeyerAI.com
