🔍 Read the full analysis: Should You Still Learn To Code? Evidence For Students, Developers And Managers on ThorstenMeyerAI.com
Get movie nights 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 narrower entry-level path, while research suggests that relying on AI during learning can weaken understanding of code.
A review published by ThorstenMeyerAI.com on November 18, 2026, finds that AI is writing a substantial share of code in some workplaces, but the evidence does not show that learning to program has become pointless. The review points to mixed productivity results, fewer entry-level software postings and an experiment in which developers learning with AI performed worse on a follow-up quiz.
Anthropic reported that, in May 2026, Claude authored more than 80% of code merged into the company’s codebase, compared with what it described as a low-single-digit share before Claude Code launched in February 2025. The company said engineers direct and review the AI-written code. It also cautioned that its reported eightfold increase in code merged per engineer measures quantity rather than quality and is “almost certainly an overstatement” of the productivity gain.
Controlled tests offer a more mixed picture of speed. In a 2025 randomized trial, nonprofit research group METR found that 16 experienced developers took 19% longer to complete 246 tasks in codebases they already knew when using the AI tools tested. The developers had expected to work faster and afterwards believed the tools had saved time. METR said a later study suggested possible speed gains, but described that evidence as “only very weak” for estimating their size.
Labor-market and learning evidence raise separate concerns. Indeed data cited by the review show that 4.5% of US software development postings were entry-level in the first quarter of 2026, while 69.3% were senior. 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 review says the largest difference involved debugging.
The Skills Behind AI-Written Code
For students and career switchers, the evidence points to a difficult but not closed entry route. A small share of advertised jobs were entry-level in Indeed’s figures, while Stanford researchers cited by the review found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it would otherwise have been. That finding concerns a defined group of workers and jobs; it does not establish that AI alone caused the change or that all young software workers face the same outcome.
For working developers, the distinction is between producing code and being accountable for software that works. The review highlights reading, testing, debugging and specifying requirements as useful capabilities when AI supplies drafts. For managers, the evidence argues against treating lines of code or self-reported speed as proof of business productivity. Quality, security, maintenance and the time spent reviewing output also matter.
Popular right now

Bose QuietComfort Wireless Noise Cancelling Headphones – Twilight Blue Limited Edition

Philips 5500 Series Fully Automatic Espresso Machine with LatteGo Milk System

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black
As an affiliate, we earn on qualifying purchases.
Why Productivity Studies Differ
The review compares studies with different tasks and settings. A 2022 experiment involving 95 professional developers found that participants using GitHub Copilot completed a small JavaScript HTTP-server task in an average of 71.17 minutes, against 160.89 minutes without it. That result cannot be directly compared with METR’s later trial, which involved experienced developers working on real tasks in large projects they already knew. Task size and codebase familiarity differed.
The evidence also comes from sources with different limitations. Anthropic’s internal figures describe a company that builds AI coding tools and uses them heavily; they do not establish the typical experience across employers. METR’s first trial tested tools available in February to June 2025, while its later study encountered developers who increasingly declined tasks without AI. Its results therefore do not provide a settled estimate of productivity in late 2026.
““More than 80% of the code we merge into Anthropic’s codebase was authored by Claude.””
— Anthropic
What the Evidence Cannot Yet Settle
The available figures do not establish whether AI has reduced the total number of software jobs, or how much of the reported weakness among younger workers is caused by AI rather than other economic factors. A low share of entry-level postings is a snapshot of advertised jobs, not a full count of hiring or employment. Anthropic’s internal code share also cannot be assumed to represent other companies.
There is no settled estimate in the cited material for how much AI speeds up experienced developers across different tasks in 2026. METR’s later results had confidence intervals that included no effect, and the researchers warned against treating them as conclusive. The review also does not establish which teaching methods can prevent the learning gaps seen in its AI-assisted experiment.
What Students and Employers Should Track
The next useful evidence will come from studies that measure completed work over time, including defects, review effort and maintenance—not just code volume or developers’ impressions of speed. Hiring data will also need to show whether entry-level opportunities recover, keep shrinking or change toward roles that combine programming with AI supervision.
Until those results are clearer, the review’s practical conclusion is to keep learning to code, while treating AI as a tool rather than a substitute for understanding. Students can practice explaining and testing generated code; developers can focus on review and debugging; managers can evaluate output quality and security alongside speed. Those are recommendations drawn from the evidence, not proof that any particular training plan guarantees better outcomes.
Key Questions
Does the evidence show that students should stop learning to code?
No. The review argues that programming remains useful, but that students should learn to understand, test and debug code, including code produced by AI. The available evidence does not establish that AI has made programming knowledge unnecessary.
Does AI make developers faster?
Results vary by task and study. METR’s 2025 trial found experienced developers took 19% longer with the AI tools tested, while a 2022 Copilot experiment found participants completed a small coding task faster. METR later said its newer evidence was too weak to give a reliable estimate of speed gains.
What did Anthropic report about AI-written code?
Anthropic said Claude authored more than 80% of code merged into its codebase in May 2026. This is a company-specific figure, and Anthropic cautioned that its eightfold code-output measure likely overstated productivity gains.
Are entry-level software jobs disappearing?
The cited Indeed data show that 4.5% of US software development postings were entry-level in the first quarter of 2026. That indicates a small share of postings in that period; it does not by itself prove that entry-level jobs have disappeared or identify the cause.
Source: ThorstenMeyerAI.com
