🔍 Read the full analysis: AI And Coding Skills: What Students, Developers And Managers Should Know 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 is writing a large share of code at some companies, but that does not establish that developers are consistently faster or that coding skills are no longer useful. Research points to a tighter entry-level job market and a risk that AI-assisted learning can weaken understanding, while demand for software roles remains.
A report published November 18, 2026, finds that AI tools now generate much of the code at some companies, but available evidence does not show a consistent productivity gain across real-world development work. The findings matter for students, developers and hiring managers: coding remains relevant, while the skills needed around AI-generated code—and access to entry-level roles—are changing.
The clearest company-level figure comes from Anthropic, which said that more than 80% of the code merged into its codebase in May 2026 was authored by Claude. The company said the share had been in the low single digits before Claude Code launched in February 2025. Anthropic also reported that a typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024, while warning that counting lines of code measures quantity rather than quality and almost certainly overstates productivity gains. These are company-reported figures, not a measure of the wider software industry.
Independent productivity evidence is mixed. In a randomized METR trial involving 16 experienced open-source developers and 246 tasks in codebases they knew, developers took 19% longer when allowed to use AI tools available from February to June 2025. They had expected AI to reduce their time and afterward believed it had made them faster. METR’s later study produced estimates pointing toward shorter task times, but the organization said the results were unreliable and offered only very weak evidence about the size of any speed-up.
Employment and education data raise separate concerns. Indeed data cited in the report show that 4.5% of US software-development job postings were entry-level in the first quarter of 2026, while 69.3% were senior. Stanford researchers found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below the level their analysis estimated it otherwise would have reached. Yet Indeed postings for US software-development roles were about 23% higher year over year in November 2026, after reaching a low in May 2025. The figures describe different measures and do not establish that AI caused the employment patterns.
Skills Beyond Generating Code
The evidence shifts the practical question from whether AI can produce code to whether people can specify, inspect and validate what it produces. Amodei’s 2025 remarks, as quoted in the report, included the point that a programmer still needs to define what is being built, how it fits with other code and whether the design is secure. That distinction matters in workplaces where code generation may be faster or more common, but errors, security weaknesses and poor fit still require human judgment.
For students and career switchers, the narrow share of entry-level postings may make it harder to gain the experience that traditionally leads to more senior work. For working developers, AI fluency alone may not replace knowledge of debugging, testing and system design. Managers face a related measurement problem: a larger volume of generated code does not by itself show that a team is delivering better software or saving time. Output, quality and productivity are not interchangeable.
An Anthropic experiment cited in the report adds a learning concern. 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 largest gap was in debugging. That result does not prove AI always impairs learning, but it suggests that delegating the work too early can leave learners with weaker understanding of how code behaves.
Products Worth Considering

LG 27G810A-B 27-inch UltraGear 4K UHD IPS Gaming Monitor

Logitech MX Master 4 Ergonomic Wireless Mouse with Haptics – Graphite

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight
As an affiliate, we earn on qualifying purchases.
Evidence From Tools and Trials
The public figures come from sources with different methods and limits. Anthropic’s internal data provide a detailed view of one company that builds and uses Claude, but its experience may not represent typical employers. 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. In February 2026, METR cited an estimate that about 4% of GitHub commits were authored by Claude Code. Those measures cover different organizations and time periods, so they should not be treated as a single industry-wide trend.
Controlled studies also test different kinds of work. A 2022 experiment with 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 instead tested real tasks in large codebases familiar to experienced contributors. The contrast underscores how results can depend on the task, the developer and the codebase—not just the AI tool.
In March 2025, Anthropic chief executive Dario Amodei predicted that AI would write 90% of code in three to six months and essentially all code in twelve months. The report notes that Anthropic’s later internal figure showed a large share of its merged code was AI-authored, though the timeline and the company’s own caveats make that claim different from proof that software development as a whole has been automated.
“More than 80% of the code we merge into Anthropic’s codebase was authored by Claude.”
— Anthropic
Limits in the Available Evidence
It remains unclear how much AI improves overall developer productivity across the industry. Anthropic’s figures are self-reported by a company with strong reasons to use its own tools, and merged lines of code do not capture software quality, maintenance costs or defects. METR’s later estimates include confidence intervals that span no effect, and its researchers said developers’ reluctance to take tasks without AI made the newer study difficult to interpret.
The job figures also do not establish why entry-level postings are scarce or why younger workers’ employment in AI-exposed jobs differs from the researchers’ estimate. The cited data show patterns, not a definitive causal link to AI. The report does not provide a single comparable measure of AI use or productivity across employers, nor does it establish how quickly hiring requirements will change.
What Employers and Learners Track
The next useful evidence will come from larger, repeated studies that measure completed work as well as time, code quality, bugs and maintenance. METR’s second study illustrates why researchers need methods that account for developers’ changing willingness to work without AI. Comparable data across different tools, teams and task types would help clarify whether productivity gains seen in one setting carry over to routine software work.
Students and career switchers can use the findings as a reason to pair AI practice with fundamentals: reading unfamiliar code, testing outputs, debugging failures and explaining design choices. Developers and managers can judge tools against their own work rather than treating code volume or self-reported speed as proof of impact. Hiring trends, learning outcomes and measured software quality will show whether AI is changing the profession’s entry routes as well as its daily tasks.
Key Questions
Does the report say students should stop learning to code?
No. It argues that coding remains useful, while emphasizing skills such as reading, testing, debugging and evaluating AI-generated code. It also cites an experiment in which junior developers who used AI to learn a library scored lower on a follow-up quiz than those who coded by hand.
Has AI been shown to make developers faster?
Not consistently. A METR randomized trial found experienced developers took 19% longer with AI tools on the tested tasks in 2025. A later study suggested possible time savings, but METR said its evidence was weak and the estimates could include no effect.
What does Anthropic’s 80% figure measure?
Anthropic said that more than 80% of code merged into its own codebase in May 2026 was authored by Claude. It is a company-specific figure, not an estimate for all software companies.
Is AI responsible for fewer entry-level software jobs?
The cited figures show a small share of entry-level postings and lower-than-estimated employment among young workers in AI-exposed jobs. They do not establish that AI caused either pattern, and the report also says software-development postings rose year over year by November 2026.
Source: ThorstenMeyerAI.com
