🔍 Read the full analysis: Should You Learn To Code When AI Can Write Code? 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 now authors a large share of code at some companies, but evidence does not show that learning to code has become pointless. Research also finds risks: AI tools may slow some experienced developers, narrow entry-level opportunities and weaken learning when users rely on generated code without understanding it.
Learning to code remains useful even as AI writes a growing share of software, according to an analysis of 2026 hiring data, workplace reports and research studies. The evidence points to a shift in what developers need to do: specify problems, understand code, test and debug it, and judge whether a design is secure—not simply type every line themselves.
Anthropic says Claude authored more than 80% of the code merged into the company’s codebase in May 2026, compared with “the low single digits” before Claude Code launched in February 2025. The company also reported that 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.
That company-specific figure does not establish how much faster software teams generally work. In a randomized METR trial conducted in early 2025, 16 experienced open-source developers completed 246 tasks in projects they knew well. They took 19% longer with AI tools, despite expecting the tools to save time and later believing they had made them faster. METR said a follow-up study offered only “very weak evidence” about the size of any improvement by early 2026.
Hiring data show pressure on newcomers, alongside signs that demand has not disappeared. Indeed data cited in the analysis put entry-level roles at 4.5% of US software-development postings in the first quarter of 2026, while senior roles made up 69.3%. Stanford researchers found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it would have been otherwise, according to their estimate. Meanwhile, US software-development postings on Indeed were about 23% higher year over year in November 2026, after reaching a low in May 2025.
Coding Skills Beyond Typing
For students and career changers, the data suggest the route into software work may be getting narrower, even if demand for developers remains. A smaller share of entry-level listings can make it harder to gain the experience employers seek. The figures do not prove that AI caused the change, but they make it important for learners to build skills they can demonstrate beyond producing code with a prompt.
For working developers and managers, AI-generated code still needs people who can identify the right problem, fit a change into an existing system and catch errors. A tool can produce code quickly without proving that the code is correct, secure or maintainable. The practical value of programming knowledge is increasingly in judging and validating output, as well as creating it.
There is also a learning trade-off. 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 gap concerned debugging. That result does not show that AI always impedes learning, but it warns that accepting generated answers without working through the underlying ideas can leave gaps.
Products Worth Considering

DJI Mini 5 Pro Fly More Combo with DJI RC 2, 1-Inch CMOS

Anker SOLIX C300 Portable Power Station, 288Wh

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue
As an affiliate, we earn on qualifying purchases.
From Code Generation to Review
The debate intensified after Anthropic chief executive Dario Amodei said in March 2025 that AI might write 90% of code within three to six months and essentially all code within a year. Anthropic’s later internal figures show substantial use at its own company, although the 90% prediction was not a general measure of the software industry and the company’s experience may not represent other employers.
Other published figures offer a different scale. Google chief executive Sundar Pichai said in October 2024 that AI generated more than a quarter of Google’s new code, 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 cover different companies, tools and periods, so they should not be treated as directly comparable industry-wide totals.
Study results also depend on the task. A 2022 experiment with 95 professional developers found participants using GitHub Copilot finished a small HTTP-server coding task 55.8% faster on average. METR’s later trial instead involved real tasks in large codebases familiar to experienced developers. The difference helps explain why a tool may speed up a contained exercise but add time when a developer must understand a project’s conventions and verify changes.
“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 Far the Shift Reaches
There is no single reliable measure of how much AI has increased software productivity across the industry. Anthropic’s code share and output figures come from a company that builds AI coding tools and uses them heavily; lines of code do not show whether software is better or easier to maintain. METR’s trials involve particular developers and tasks, and its later study could not establish a clear effect size.
The job figures also do not establish that AI alone caused weaker employment among young workers or fewer entry-level postings. The evidence summarized here does not show how those trends will develop, whether new roles will offset lost opportunities, or which specific coding skills employers will prioritize. The learning experiment points to a possible risk, not a rule that every learner using AI will retain less.
What Learners Should Track
For people deciding what to study, the available evidence favors learning programming fundamentals while practicing with AI tools rather than treating coding as either obsolete or unchanged. Learners can use generated code as material to inspect: explain what it does, test edge cases, find and fix bugs, and check for security problems. Employers and educators will also need to determine how to assess those skills when tools can produce working-looking code.
The next useful signals will be better-controlled studies of current AI tools, comparable productivity measures across different kinds of software work, and continued hiring data that separate entry-level from senior roles. Until those arrive, the amount of code AI writes is not a clear measure of how many developers will be needed or what a new programmer’s career prospects will be.
Key Questions
Is learning to code still worth it?
The evidence does not support treating coding as a waste of time. It suggests learners should combine programming fundamentals with the ability to specify tasks, review AI-generated code, test it and debug it.
Does AI make software developers faster?
Not consistently in the available studies. METR’s early-2025 trial found experienced developers took 19% longer with AI on real tasks in familiar codebases, while a 2022 experiment found faster completion of a small coding exercise. Results depend on the task and study conditions.
Are entry-level software jobs disappearing?
The cited data show entry-level roles were 4.5% of US software-development postings on Indeed in the first quarter of 2026, and Stanford researchers estimated weaker employment among young people in AI-exposed jobs. These figures do not prove AI alone caused the changes or establish what will happen next.
Can using AI to learn programming affect understanding?
In one Anthropic experiment, 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 result suggests learners should actively explain, test and debug generated code rather than rely on answers they do not understand.
Source: ThorstenMeyerAI.com
