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

🔍 Read the full analysis: Students, Developers And Managers: Should You Still Learn To 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 producing a substantial share of code at some companies, but evidence that it consistently makes developers faster is mixed. Hiring data suggest fewer entry-level openings, while demand for software developers has recently risen; learning to code still matters, though the skills taught may need to change.

AI is writing a large share of some companies’ code, but available evidence does not show that learning to program has become unnecessary. A review published by ThorstenMeyerAI.com on November 18 weighs company disclosures, employment data and developer studies; it points to a changing job market and a greater need for people who can specify, test, debug and review software.

Anthropic reported that more than 80% of the code merged into its codebase in May 2026 was authored by Claude, compared with “the low single digits” before Claude Code launched in February 2025. The company also said 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.

Evidence on whether AI makes developers faster is less settled. 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 the AI tools tested, even though they had expected a time saving. METR said a later study gave only weak evidence about the size of any speed-up by early 2026.

Hiring figures show pressure on entry-level applicants, but not an across-the-board collapse in demand. Indeed data cited in the review put entry-level roles at 4.5% of US software development postings in the first quarter of 2026, while senior roles accounted for 69.3%. The review also reports that US software development postings were about 23% higher year over year in November 2026, after reaching a low in May 2025.

At a glance
analysisWhen: Published November 18, 2026; the cited…
The developmentA 2026 review of company figures, job listings, experiments and surveys finds that AI is changing how software is written, without establishing that coding skills are no longer needed.

Coding Skills as Entry Jobs Narrow

For students and career changers, the figures describe a difficult distinction: software work remains in demand, but the first rung may be harder to reach. If employers expect AI to handle routine implementation, junior candidates may face greater pressure to demonstrate that they can understand existing systems, find errors and make sound technical decisions. The cited job-posting data describe openings, not hires or the reasons employers set their experience requirements.

For working developers, AI use does not remove the need to understand the output. Anthropic’s description is of engineers directing and reviewing code written by Claude, not handing over all responsibility. Managers, meanwhile, should not treat higher code volume as proof of equivalent gains in quality, delivery time or security. Those outcomes require separate measurement.

The review also flags a learning risk. 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 largest gap was in debugging. That result is limited to the experiment described; it does not establish that every use of AI weakens learning. It does show why students may need practice working through code rather than relying on generated answers alone.

Products Worth Considering

As an affiliate, we earn on qualifying purchases.

From Code Generation to Review

In March 2025, Anthropic chief executive Dario Amodei predicted that AI would write 90% of code within three to six months and “essentially all” of it within a year. The company’s later figures indicate extensive use inside Anthropic, but they do not verify that prediction across the software industry. Anthropic is both a developer of AI coding tools and a company with strong incentives to use them, so its internal figures should not be read as a representative industry average.

Other figures in the review offer different snapshots. Google chief executive Sundar Pichai said in October 2024 that AI generated more than a quarter of the company’s new code, which engineers then reviewed and accepted. METR cited an estimate in February 2026 that Claude Code had authored about 4% of GitHub commits. These figures cover different organizations, periods and measures, and cannot be compared as a single industry-wide trend.

Productivity studies also test different kinds of work. A 2022 experiment with 95 professional developers found that participants using GitHub Copilot completed a small HTTP-server task in JavaScript 55.8% faster on average. METR’s later trial instead involved real tasks in large codebases familiar to experienced developers. The contrast matters: results from a bounded coding exercise may not predict performance on work involving a project’s conventions, dependencies and review requirements.

“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 in March 2025

What the Evidence Cannot Settle

It remains unclear how much AI improves overall developer productivity across different tasks, experience levels and workplaces. METR’s early-2026 estimates had confidence intervals that included no effect; the group described its evidence for a larger speed-up as very weak. Tools and work practices have also changed since the early-2025 trial, limiting how directly its result applies today.

The employment figures do not establish that AI caused the low share of entry-level postings or the reported employment decline among 22- to 25-year-olds in AI-exposed jobs. The review cites Stanford researchers’ estimate that employment for that group was 19% below where it would have been, but the supplied material does not establish the causes or provide details needed to assess the full method here. Nor do posting shares show how many roles were filled, how many applicants competed for them, or how employers’ requirements changed.

Anthropic’s figures are self-reported and specific to its own organization. The source material does not provide a comparable, independently verified measure of code quality, security outcomes or productivity across a broad range of employers. It is also not yet clear whether AI-assisted learning produces different results over longer periods or when teaching methods explicitly require students to explain and test generated code.

Watch Hiring and Learning Outcomes

The next useful evidence will come from repeated productivity studies that test current AI tools on varied, real-world tasks and report time, quality and error rates together. METR said its newer work was difficult to interpret partly because some developers were reluctant to take on tasks without AI; clearer study designs will be needed to separate tool effects from changes in what developers choose to do.

Readers can also watch whether job-posting trends continue beyond November 2026 and whether entry-level hiring changes alongside overall demand. The cited figures are snapshots and do not establish what employers will hire for next. For schools, training programs and managers, the practical question is whether learners can both use AI tools and demonstrate independent understanding through testing, debugging and code review.

On the evidence presented so far, the answer to whether people should learn to code is yes, but coding education should not stop at writing syntax. How much curriculum or hiring practice changes will depend on better evidence about the quality of AI-assisted work, its effects on learning and the kinds of software roles employers continue to create.

Key Questions

Does AI make learning to code pointless?

No. The cited evidence shows that AI can generate substantial amounts of code in some workplaces, but people are still needed to specify requirements, test results, debug problems and judge design decisions. The evidence does not show that programming knowledge is no longer useful.

Are developers faster when they use AI?

Results vary by study and task. In METR’s early-2025 trial, experienced developers took 19% longer using the tools tested, while a 2022 Copilot experiment found participants completed a small coding task 55.8% faster. METR said evidence from its later study was too weak to establish the size of a newer speed-up.

Are entry-level software jobs disappearing?

The review reports that entry-level positions 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 narrow entry-level share in that dataset, but do not prove that jobs are disappearing or establish why the share is low.

What should students focus on learning?

The evidence in the review supports learning to read and review code, write clear specifications, test and debug programs, and assess design and security. Students should also build the ability to explain and verify AI-generated code rather than relying on output they do not understand.

Source: ThorstenMeyerAI.com

You May Also Like

How OpenAI’s Price Cuts For GPT‑6 Sol And Luna Preserve Benchmark Stability

OpenAI reduces GPT‑6 Sol and Luna prices by 50%, keeping performance stable while enhancing cost efficiency, impacting AI deployment strategies.

NVIDIA Runs Simulation Of Fault-Tolerant Quantum Computing At Jülich Supercomputing Centre

NVIDIA says its 8,192-GPU JUPITER simulation modeled fault-tolerant quantum computing with realistic hardware noise and tested magic-state cultivation.

The Astra Controversy: Crossing Boundaries And Remaining Gated

OpenAI’s Astra model has achieved ‘Critical’ cybersecurity capability status, raising concerns about safety, safeguards, and responsible deployment amid ongoing management efforts.

Claude Fable 5.1 Dominates The AI Index — Insights From The Cost Line

Fable 5.1 leads the AI Intelligence Index with a score of 66, outperforming competitors but at roughly 20% higher per-task cost due to increased verbosity.