AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Should You Learn Programming When AI Can Generate Code? on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

AI is writing much of the code at some companies, but evidence that it consistently makes developers faster is mixed. Hiring data show a tougher entry-level market, while a learning experiment found that junior developers using AI performed worse on a follow-up quiz than peers who coded by hand. The evidence supports learning programming alongside skills in specifying, testing, debugging and reviewing code.

AI is generating a growing share of software code, but the available evidence does not show that learning programming has become unnecessary. Data cited in a November 18 analysis by ThorstenMeyerAI.com point to a more complicated picture: AI use is extensive at Anthropic, productivity results vary by task, and entry-level software hiring remains difficult even as job postings have recently risen.

Anthropic says that in May 2026, Claude authored more than 80% of the code the company merged into its codebase, 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 the productivity gain.

Independent evidence on speed is less conclusive. In a 2025 randomized trial, the nonprofit METR assigned 16 experienced open-source developers 246 tasks in projects they knew well. Developers using the available AI tools took 19% longer to complete the tasks, despite expecting AI to save time and later believing it had made them faster. METR said in February 2026 that newer results suggest developers may be getting more help from AI, but described those results as very weak evidence about the size of the effect.

The employment signals also point in different directions. Indeed data cited by ThorstenMeyerAI.com show that just 4.5% of US software-development 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 where it would otherwise have been. At the same time, Indeed’s US software-development postings were up about 23% year over year in November 2026, after reaching a low in May 2025. These figures describe different measures and do not establish that AI alone caused hiring changes.

At a glance
analysisWhen: Published November 18, 2026; evidence a…
The developmentA 2026 review of company data, hiring figures, controlled studies and developer surveys finds that AI can generate substantial amounts of code, but does not establish that programming skills are no longer useful.

Programming Skills Beyond Code Generation

For students and career switchers, the evidence suggests a narrower entry point, not a settled case that coding careers are disappearing. If employers expect new hires to work with AI tools, applicants may need to show they can do more than produce a working snippet: they may also need to explain requirements, find defects and judge whether proposed code fits a larger system.

How people learn matters too. In an Anthropic experiment described in the source analysis, 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. Debugging produced the largest gap. This result does not prove that AI always harms learning; it does show that having a tool generate code is not the same as understanding it.

For working developers and managers, the distinction between more code and more useful output matters. A larger volume of generated code may create review, testing and maintenance work rather than an equivalent increase in completed, reliable software. Hiring and training decisions based only on code-generation speed may miss the skills needed to catch mistakes and maintain systems.

Products Worth Considering

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

As an affiliate, we earn on qualifying purchases.

Why Code Output Is Not Productivity

Anthropic’s figures are unusually direct, but they come from a company that builds Claude and uses its own coding tools heavily. They should not be treated as a measure of how every software team works. In October 2024, Google chief executive Sundar Pichai said more than a quarter of new code at Google was generated by AI and then reviewed and accepted by engineers. A February 2026 estimate cited by METR put Claude Code at about 4% of GitHub commits. These figures use different sources and measures, so they are not directly comparable.

Controlled studies also vary with the work being tested. A 2022 experiment with 95 professional developers found that participants using GitHub Copilot completed a short JavaScript HTTP-server task 55.8% faster on average. METR’s later trial instead involved real tasks in large, familiar codebases. The contrast suggests results from a bounded coding exercise may not predict performance on complex work with existing rules, dependencies and review requirements.

In a March 2025 speech, Anthropic chief executive Dario Amodei forecast that AI would write 90% of code within three to six months and essentially all of it within a year. He also said programmers would still need to specify what was being built, how it fit with other code and whether the design was secure. The forecast is a claim about the expected pace of change, not evidence that those timelines were met across the industry.

“Almost certainly an overstatement of the true productivity gain.”

— Anthropic, discussing its eightfold code-output figure

Hiring and Productivity Remain Unsettled

The figures do not establish how much AI caused the changes in software employment. The cited Stanford finding concerns young workers in AI-exposed jobs, while Indeed’s figures track the share and number of postings; neither measure alone identifies the reasons behind hiring patterns. A rise in postings does not show that entry-level opportunities have recovered.

The productivity evidence also has limits. METR’s first trial involved a small group of experienced developers working in projects they already knew, and its later estimates had confidence intervals that included no effect. Anthropic’s internal code-output figures describe one AI company and count lines merged, not whether the code was correct, valuable or costly to maintain. The source analysis does not provide enough detail to determine how well the learning experiment applies to other learners, tools or teaching methods.

It remains unclear how these results will change as tools and workplace practices develop, which programming roles will be most affected, and whether employers will create different entry-level pathways. The evidence supports neither a guarantee of future demand nor the claim that learning to program is now pointless.

Watch Hiring and Learning Results

Readers weighing a programming course or career change should watch for updated hiring data and studies that measure completed work, not just generated code. Future research can help establish whether newer AI systems improve performance on realistic software tasks and whether those gains hold across experience levels and project types.

For now, the evidence supports learning the fundamentals while using AI as a tool rather than a substitute for understanding. Practice writing clear specifications, reading generated code, testing behavior, debugging failures and reviewing security. Those skills are directly relevant to the responsibilities described in the source material, although the available evidence does not guarantee which abilities employers will prioritize.

Employers and educators will also need to evaluate whether AI-assisted learning builds durable skills. The next useful signal will be evidence that tracks learners and developers over time, alongside job postings and hiring outcomes. Until then, the practical answer is to learn programming, including how to judge code an AI produces.

Key Questions

Does the evidence show that AI makes all programmers faster?

No. In a 2025 METR randomized trial, experienced developers took 19% longer on selected real tasks when using the AI tools available in the study. METR’s later results suggest the effect may have improved, but it said those data provide only weak evidence about the size of any speed-up.

Should beginners still learn programming?

The cited evidence does not show that programming has become unnecessary. It suggests learners should build an understanding of code and practice testing, debugging, specifying requirements and reviewing AI-generated work, rather than relying only on code generation.

Is entry-level software hiring disappearing?

The source analysis reports that entry-level roles made up 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 2026. Those statistics do not establish that entry-level hiring is disappearing or identify AI as the cause of the pattern.

Can using AI make it harder to learn coding?

An Anthropic experiment found that mostly junior developers who used AI to learn a new library scored 50% on a follow-up quiz, compared with 67% for those who coded by hand, with the largest gap in debugging. The finding is a warning about one learning setup, not proof that every use of AI reduces learning.

Source: ThorstenMeyerAI.com

You May Also Like

The Simple Science Behind Deep Learning And AI At Work

Deep learning trains networks by measuring errors and adjusting weights. The method explains AI’s strengths, limits and workplace role.

The 3‑2‑1 Backup Rule: How to Protect Your Files Without Overthinking It

Learn how the 3-2-1 backup rule can effortlessly protect your files and why mastering it is essential for avoiding data loss.

Best Low-Noise PC Cases for Airflow and Sound Dampening

Explore top PC cases balancing airflow and sound dampening, ideal for high-power workstations and quiet builds, with expert insights and options.

Thrymvault In The Mix: My AI Agents’ Shared Desk

Thorsten Meyer describes Thrymvault, a self-hosted workspace designed to let editors track and steer AI agents’ newsroom work.