🔍 Read the full analysis: Should Everyone Learn To Code As AI Becomes More Capable? on ThorstenMeyerAI.com
Get the latest gadgets 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 now authoring a large share of code at some companies, while studies show its effect on developer speed and employment is mixed. The evidence supports learning to code, but with greater emphasis on specifying, reviewing, testing and debugging software—not just writing it by hand.
AI is writing more software code, but evidence on whether it makes developers faster or eliminates the need for programming skills remains mixed, according to a review published November 18 by ThorstenMeyerAI.com. The review argues that students and working developers should still learn to code, while placing more weight on defining requirements, reviewing output, testing and debugging.
The clearest public measure comes from Anthropic, which reported that in May 2026 more than 80% of code merged into its own codebase 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 its typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024, with engineers directing and reviewing code written by Claude.
Those figures do not establish an equivalent increase in productivity. Anthropic cautioned that lines of code measure quantity rather than quality and said the eightfold figure was “almost certainly an overstatement” of the actual productivity gain. Its March 2026 employee poll found a median estimate of roughly four times as much output with AI; the company said it expected the true increase to be lower. Anthropic is also a developer of AI coding tools, and its own results may not represent other employers.
Controlled studies and hiring data offer a less settled picture. In a 2025 randomized METR trial, 16 experienced open-source developers took 19% longer on real tasks when using AI tools, despite believing the tools had made them faster. METR said a later study suggested developers may have gained speed by early 2026, but called that evidence very weak. Meanwhile, Indeed data cited in the review show that entry-level roles made up 4.5% of US software-development postings in the first quarter of 2026, while senior roles accounted for 69.3%. US software-development postings were up about 23% year over year in November, after reaching a low in May 2025.
Coding Skills Beyond Writing Code
The immediate question for learners is not simply whether AI can generate code. It is whether people can tell the system what to build, recognize when its output is wrong, and make software work safely within a larger system. Those abilities matter to students, career switchers and employers as AI takes on more of the typing.
A separate Anthropic experiment points to a trade-off in how people acquire those skills. Mostly junior developers who used AI while learning a new library scored 50% on a follow-up quiz, compared with 67% among participants who coded by hand. The largest gap concerned debugging. The result does not prove AI always weakens learning, but it suggests that delegating practice can leave learners less prepared to diagnose mistakes.
For employers, the data support caution in both hiring and training decisions. A smaller share of entry-level listings may make the first step into software work harder, but the figures do not show that demand has disappeared. And code volume alone cannot show whether a team is delivering more reliable software or completing work faster. Review, testing and security judgment remain relevant even when a model writes much of the code.
Products Worth Considering

Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black

Apple Watch SE 3 GPS + Cellular 40mm Smartwatch

LG 27GR83Q-B 27-inch Ultragear QHD IPS Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Copilot Tests to Claude Code
Predictions about AI-written code have come from both company leaders and measured company practice. On March 10, 2025, Anthropic chief executive Dario Amodei told the Council on Foreign Relations that AI would be writing 90% of code in three to six months and essentially all of it within twelve months. The subsequent May 2026 figure from Anthropic—more than 80% of code merged at the company authored by Claude—shows substantial use there, but does not verify the full prediction across the software industry.
Results have varied by task. A 2022 study of 95 professional developers found that participants using GitHub Copilot completed a small JavaScript HTTP-server task 55.8% faster on average than those working without it. METR’s later trial involved experienced developers working on real tasks in large codebases they already knew. The difference in task size and setting makes the results difficult to treat as direct contradictions: assistance on a contained exercise may not translate into faster work inside a complex project.
Other figures also have limits. Google chief executive 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. These measurements cover different companies, tools, time periods and definitions of authorship, so they do not form a single industry-wide rate.
“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
Productivity and Hiring Remain Unsettled
The available findings do not settle whether AI makes software developers more productive overall. METR’s early-2025 controlled trial measured slower completion, while its later study produced estimates consistent with modest time savings but confidence intervals that included no effect. METR also said some developers were reluctant to take on tasks without AI, complicating comparisons. Tools and workflows have changed since the first trial, so its result should not be treated as a permanent measure of current performance.
Employment figures require similar care. Stanford researchers cited in the review found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it would have been, but that finding does not by itself show AI caused the difference or identify the effect on software jobs alone. The decline in entry-level postings and the later rise in overall software-development postings describe different measures. The sources do not establish how much AI, economic conditions or other changes explain either trend.
Nor do the company coding figures reveal how often generated code requires substantial rework, how its quality compares across projects, or what share of software work remains outside code production. The evidence supports a changing division of tasks, not a settled forecast that programmers will or will not be needed.
Watch Hiring and Better Trials
The next useful evidence will come from longer, controlled studies that track task completion, code quality and rework across different experience levels and software projects. METR’s account warns that its later experiment could not give a reliable estimate of the productivity effect, leaving that question open.
Readers can also track whether software-development postings continue to recover and whether the share of entry-level openings changes. The November 2026 review provides a snapshot, not a forecast. Companies’ claims about code authorship and output should be read alongside independent measures of quality and completed work.
For people deciding what to study now, the available evidence points to learning programming fundamentals while practicing with AI tools rather than replacing fundamentals with prompting alone. Whether that combination improves learning and job prospects at scale remains an open question.
Key Questions
Should beginners still learn to code?
The evidence does not show that learning to code is a waste of time. It suggests learners should build fundamentals and practice specifying tasks, reviewing generated code, testing and debugging it.
Does AI make software developers faster?
Results vary. METR’s randomized 2025 trial found experienced developers took 19% longer with AI on its tasks. A later study suggested possible time savings, but METR said the evidence was very weak and included no clear effect.
Is AI already writing most code?
Anthropic reported that Claude authored more than 80% of code merged into Anthropic’s codebase in May 2026. That is a company-specific figure and does not establish the share across the software industry.
Are entry-level software jobs disappearing?
Indeed data cited in the review show entry-level roles were 4.5% of US software-development postings in the first quarter of 2026, while overall postings were up about 23% year over year in November. Those figures do not establish that AI caused the entry-level share or that demand has disappeared.
Source: ThorstenMeyerAI.com
