🔍 Read the full analysis: Do Developers Still Need To Learn To Code? What The Evidence Says 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 that it consistently makes developers faster is mixed. Hiring data point to a tighter entry-level market, while experiments suggest AI-assisted learning can leave gaps in debugging and understanding.
AI is writing a growing share of software code, but evidence available in 2026 does not show that developers can safely stop learning programming. Company figures, controlled tests and labor-market data instead point to a changing skill mix: developers may delegate more typing, while still needing to specify work, review results, debug failures and judge security.
Anthropic reported that more than 80% of the code merged into its codebase in May 2026 was authored by Claude, compared with a share in the low single digits before Claude Code launched in February 2025. The company said its typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024. It cautioned that lines of code measure quantity rather than quality and that the eightfold increase is “almost certainly an overstatement” of productivity gains.
Independent evidence on speed is less decisive. In a randomized METR study, 16 experienced developers completed 246 tasks in familiar open-source projects. With the AI tools available during the February-to-June 2025 trial, they took 19% longer, despite expecting a time saving and later believing they had worked faster. METR’s follow-up work suggested developers may have gained more from AI by early 2026, but the organization described that newer evidence as very weak and said its estimates could not establish the size of any speed-up.
The labor market presents a mixed picture. Indeed data 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 estimated employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it otherwise would have been. But Indeed postings for US software-development roles were up about 23% year over year in November 2026, following a low in May 2025. Those figures describe different measures and do not, by themselves, show that AI caused changes in hiring.
Programming Skills Beyond Code Generation
For students and career changers, the evidence points to a more difficult entry route, not proof that programming knowledge has become unnecessary. If employers expect new hires to work with AI-generated code, the ability to explain a task, spot an incorrect result and trace a bug can matter more, not less. A person who can produce code with a tool but cannot tell whether it works may struggle to take responsibility for software in production.
That distinction also matters to employers deciding whom to hire and train. Anthropic’s figures show that AI can generate code at high volume within a company built around AI tools. They do not establish that the same output is useful, safe or economical across other organizations. The company itself says its code-volume measure overstates productivity gains. Managers need to evaluate outcomes such as correct, maintainable software and effective delivery, rather than treating lines produced as a direct measure of value.
Learning methods are part of the issue. 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 developers who coded by hand; the largest gap was in debugging. The result does not prove that AI always weakens learning, but it cautions against assuming that completing a task with assistance means a learner has understood the underlying concepts.
Products Worth Considering

Ledger Nano X – Classic Crypto Wallet with Bluetooth

Trezor Safe 7 Crypto Hardware Wallet with Bluetooth and Color Touchscreen

LG 27G810A-B 27-inch UltraGear 4K UHD IPS Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From AI Predictions to Measured Results
On March 10, 2025, Anthropic chief executive Dario Amodei told a Council on Foreign Relations audience that AI would be writing 90% of code within three to six months and essentially all of it within 12 months. He also said programmers would still need to specify what was being built, how it fitted with other code, and whether the design was secure. Anthropic’s later internal figures give one measure of rapid adoption at his company, but they do not validate the prediction across the software industry.
Other published estimates offer narrower snapshots. 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 Claude Code authored approximately 4% of GitHub commits. These measures cover different companies, tools and periods, so they should not be treated as a single industry-wide rate.
Productivity experiments have also produced different results. A 2022 test involving 95 professional developers found that participants using GitHub Copilot finished a small JavaScript HTTP-server task in an average of 71.17 minutes, compared with 160.89 minutes without it. That task differed from METR’s work on real tasks in large codebases familiar to developers. The contrast is a reminder that results can depend on the work being measured, the developer’s experience and the tools available.
“The programmer still needs to specify what you’re building, 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
What the Evidence Cannot Settle
The figures do not establish a single, reliable productivity effect for all developers. Anthropic’s code-share and output numbers come from the company that makes Claude, and its own report warns that code volume is not a direct measure of quality or productivity. METR’s early-2025 trial was relatively small and focused on experienced developers working in projects they already knew; its later study had participation problems and estimates whose confidence intervals included no effect.
It is also unclear how much AI adoption caused the changes in entry-level employment or job postings. The figures describe hiring patterns, not a controlled test of what drove them. Nor do the available findings settle how durable AI-assisted learning gaps are, whether different teaching approaches can reduce them, or how results vary across programming languages and kinds of work. Claims about the future of coding jobs should be treated as forecasts, not confirmed outcomes.
The Next Tests for Developers
The next useful evidence will need to measure more than how much code a tool generates. Larger studies across different experience levels and real work settings could track whether AI changes delivery time, defect rates, security problems, maintenance costs and what developers retain after learning. Updated labor-market data may also clarify whether the current shortage of entry-level postings persists as overall software hiring changes.
For now, the evidence supports learning programming alongside using AI tools: practice reading unfamiliar code, writing tests, debugging, defining requirements and reviewing security, as well as producing code. These are findings about skills and workplace evidence, not a guarantee of employment or a prescription for every role. How much coding developers need to do themselves will vary by task and employer, and the measured effects remain unsettled.
Key Questions
Does the evidence show that developers no longer need to learn to code?
No. The evidence shows that AI can generate a substantial amount of code in some settings, but developers still need skills to specify, test, debug and review the result. It does not establish that programming knowledge is obsolete.
Are developers faster when they use AI?
Results vary. In METR’s early-2025 randomized trial, experienced developers took 19% longer with AI on the tasks studied. METR later said developers were likely getting more speed from newer tools, but described the evidence on the size of that effect as very weak.
Is it harder to get an entry-level software job?
Indeed reported that entry-level roles made up 4.5% of US software-development postings in the first quarter of 2026, while senior roles made up 69.3%. Those figures indicate a limited entry-level share of postings, but do not prove AI caused it or describe every employer.
Can using AI to learn programming reduce understanding?
One Anthropic experiment found that mostly junior developers who used AI while learning a new library scored 50% on a follow-up quiz, versus 67% for those who coded by hand. That result raises a concern, especially about debugging, but does not establish that all AI-assisted learning has the same effect.
Source: ThorstenMeyerAI.com
