🔍 Read the full analysis: Should Students Still Learn Coding As AI Advances? on ThorstenMeyerAI.com
Get hardware and tech essentials 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 tools are writing a growing share of code at some companies, but research does not show that programming skills are no longer useful. Evidence points to a changing role for developers and a need for students to learn how to specify, test, debug and evaluate code—not just produce it.
Students should still learn to code, but the skills that matter are shifting as AI tools take on more code production, according to a November 18 review of company disclosures, job-posting data and research experiments. Anthropic says Claude authored more than 80% of code merged into its codebase in May 2026; other evidence, however, does not establish that AI consistently makes developers faster or that programming knowledge is no longer needed.
Anthropic reported that the share of its merged code authored by Claude rose from the low single digits before February 2025 to more than 80% in May 2026. It also said its typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024. The company cautioned that lines of code measure quantity rather than quality and that the eightfold figure is almost certainly an overstatement of the productivity gain. Anthropic is an AI developer using its own tools, so its figures should not be treated as representative of all employers.
Independent productivity evidence is less settled. In a 2025 randomized trial, the nonprofit METR assigned 16 experienced open-source developers 246 tasks in projects they knew well. Developers using AI took 19% longer to complete tasks, even though they believed the tools had made them 20% faster. A separate 2022 experiment involving 95 developers found that participants using GitHub Copilot finished a small JavaScript server task 55.8% faster. The studies tested different kinds of work, making their results difficult to compare directly.
METR said its later research produced only very weak evidence about how much AI sped up developers by early 2026. Meanwhile, Indeed data show that just 4.5% of U.S. software-development job postings were entry-level in the first quarter of 2026, compared with 69.3% categorized as senior. Yet postings were about 23% higher year over year in November 2026 after reaching a low in May 2025. These figures describe postings, not hires, and do not by themselves establish that AI caused the narrow entry-level share.
The Skills Behind AI-Assisted Coding
The evidence points to a distinction between producing code and understanding what it should do. Developers still need to define requirements, fit new code into existing systems, test its behavior, find faults and judge whether a design is secure. These tasks become more consequential when a tool can generate code quickly but cannot be assumed to understand every project constraint or risk.
That distinction matters for students choosing what to study and for employers deciding whom to train. If entry-level openings are a small share of postings, new programmers may face a harder route into the field. But the available figures do not show that those jobs have disappeared, or that coding education has become pointless. They support a more limited conclusion: learning to write code remains useful, while learning to read, debug and evaluate it is also important.
There is a learning risk when AI supplies answers before students build their own understanding. In an Anthropic experiment, mostly junior developers who used AI while learning a new library scored 50% on a follow-up quiz, compared with 67% among those who coded by hand. The result suggests that speed or task completion alone may not show whether a learner has retained the concepts. It does not establish that every use of AI harms learning.
Products Worth Considering

Logitech MX Master 3S Bluetooth Wireless Mouse, Graphite

Ledger Nano X Crypto Hardware Wallet – Bluetooth (Black + Billfodl)

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight
As an affiliate, we earn on qualifying purchases.
What the Studies Actually Tested
In March 2025, Anthropic CEO Dario Amodei told an audience at the Council on Foreign Relations that AI would write 90% of code within three to six months and essentially all of it within 12 months. His forecast was a claim about what he expected, not a verified prediction that applied across the industry. Anthropic’s later figures offer evidence of heavy AI use at his company, but not a measure of all software development.
Amodei also said the programmer would still need to specify what was being built, how it fit with other code and whether the design was secure. That qualification aligns with the distinction in the research between code generation and the broader work of software development. The task matters: a short, self-contained programming exercise can produce different results from work inside a large, familiar codebase with established rules.
Other published figures provide points of comparison, not a complete industry count. Google CEO 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 Claude Code authored about 4% of GitHub commits. Different dates, measures and organizations mean these figures cannot be combined into a single share of code written by AI.
“The programmer still needs to specify”
— Dario Amodei, Anthropic CEO, speaking at the Council on Foreign Relations in March 2025
Where the Evidence Still Falls Short
There is no single, comparable measure here of AI’s effect on software work across employers. Company disclosures describe particular organizations; experiments test specific tasks and groups; job-posting counts do not show how many people were hired or why employers set their requirements. Anthropic’s code-share and output figures also come from a company that builds AI coding tools, and its own caveats warn against treating code volume as productivity.
The effect on learners is also not settled by one quiz experiment. The reported score gap is a finding from a particular exercise with mostly junior developers, not proof that all AI-assisted instruction reduces learning. It remains unclear which teaching approaches let students benefit from AI while retaining the ability to solve problems independently.
Labor-market changes require similar care. Stanford researchers found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it would have been, but that estimate does not, on its own, establish that AI caused the difference. The supplied figures do not explain how much of the smaller entry-level share reflects AI, economic conditions, hiring practices or other factors.
What Students and Schools Can Track
Further evidence will depend on studies that measure real-world tasks with newer tools, report uncertainty clearly and distinguish code output from quality, review time and overall completion time. Employers’ hiring data may also clarify whether entry-level opportunities are shrinking, recovering or changing in the skills they require. The current posting figures show a low entry-level share alongside a year-over-year rise in total software-development postings, but they do not settle what comes next.
For schools, the immediate question is how to teach programming so students can use AI without relying on it to do the thinking. Courses can test whether learners can explain generated code, identify bugs, verify behavior and justify design choices, as well as write programs themselves. The research cited here does not establish a single best curriculum, but it gives educators a reason to measure understanding rather than code produced alone.
The practical conclusion is not to abandon coding education. Students can learn programming fundamentals while practicing how to direct, inspect and test AI-generated work. How much time should go to each skill will depend on new evidence about learning outcomes and actual workplace tasks.
Key Questions
Should students still learn programming?
Yes. The evidence shows AI can generate substantial amounts of code in some settings, not that students no longer need to understand programming. Reading, testing, debugging and evaluating code remain relevant skills.
Does AI make software developers faster?
Results vary by task and study. A 2025 METR trial found experienced developers took 19% longer on selected real-world tasks while using AI, while a 2022 experiment found faster completion on a small coding exercise. METR later described its evidence of larger gains by early 2026 as very weak.
Are entry-level software jobs disappearing?
Indeed data show entry-level roles made up 4.5% of U.S. software-development postings in the first quarter of 2026, while senior roles accounted for 69.3%. Those are shares of postings, not proof of why the mix changed or how many jobs were filled. Total postings were about 23% higher year over year in November 2026.
Can using AI to learn code affect understanding?
One Anthropic experiment found mostly junior developers who used AI to learn a library scored 50% on a follow-up quiz, compared with 67% for those who coded by hand. That result raises a concern about retained understanding, but it does not establish that every form of AI-assisted learning has the same effect.
Source: ThorstenMeyerAI.com
