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

🔍 Read the full analysis: Does AI Make Coding Knowledge Less Important? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

AI writes a large share of code at some companies, but that does not establish that developers are consistently faster or that coding knowledge is no longer needed. Research and labor-market data suggest programming work is changing, with particular uncertainty around entry-level opportunities and how AI affects learning.

AI is writing much of the code at some companies, but evidence available in 2026 does not show that coding knowledge is obsolete or that AI consistently makes developers faster. Anthropic reported that Claude authored more than 80% of code merged into its codebase in May 2026; experiments, hiring data and learning research point to a more mixed picture, with developers still expected to direct, test and judge the work.

Anthropic’s own figures show how quickly AI use grew inside the company. It said Claude authored code in the “low single digits” before Claude Code launched in February 2025, compared with more than 80% of merged code in May 2026. 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 said the eightfold figure is “almost certainly an overstatement” of the productivity gain. Its March 2026 employee poll put the median estimate of AI-assisted output at about four times as much, which Anthropic said may itself overstate the real uplift.

Measured speed is less settled. In a randomized METR trial involving 16 experienced developers and 246 tasks in codebases they already knew, use of AI tools available from February to June 2025 increased task completion time by 19%. Participants had expected a 24% time saving and afterwards believed they had saved 20%. METR’s later study produced estimates pointing to faster work, but confidence intervals included no effect. The group said those data are only very weak evidence about the size of any improvement by early 2026.

The labor-market evidence is also mixed. Indeed data cited in the report put entry-level roles at 4.5% of US software development postings in the first quarter of 2026, while senior roles made up 69.3%. Stanford researchers found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it otherwise would have been, though that finding does not by itself establish AI as the cause. Meanwhile, US software development postings on Indeed were about 23% higher year over year in November 2026, after reaching a low in May 2025.

At a glance
reportWhen: Published November 18, 2026; evidence s…
The developmentA November 2026 review of company disclosures, experiments, job data and developer surveys finds that AI is changing coding work, while evidence does not show that programming knowledge has become dispensable.

Why Core Coding Skills Still Matter

The change is less about whether code can be produced and more about who can tell whether it is useful, correct and safe. AI can generate a draft, but a developer must translate a request into precise requirements, understand how a change fits an existing system, test its behavior and find defects. Those responsibilities matter to employers and to anyone relying on software, especially where errors can create security or reliability risks.

Learning itself may also be affected. 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; debugging showed the largest gap. The result does not prove AI always harms learning, but it suggests that getting a working answer can differ from acquiring the knowledge needed to diagnose a failure later.

For people entering the field, a narrower share of junior postings could make the first job harder to find, even as overall postings recover. For working developers, the evidence supports learning to review, test, debug and specify code alongside learning to write it. That is an interpretation of the findings, not proof that any single skill set guarantees employment.

Vitamix 5200 Professional-Grade Blender with 64 oz Container, Black

Vitamix 5200 Professional-Grade Blender with 64 oz Container, Black

Ninja Foodi 6-in-1 DualZone FlexBasket Air Fryer, 7-Qt MegaZone, Black, DZ071

Ninja Foodi 6-in-1 DualZone FlexBasket Air Fryer, 7-Qt MegaZone, Black, DZ071

As an affiliate, we earn on qualifying purchases.

From Code Generation to Review

The debate intensified after Anthropic chief executive Dario Amodei said at the Council on Foreign Relations on March 10, 2025, that AI would write “90 percent of the code” within three to six months and “essentially all of the code” within twelve. The source material reports that the first projection came close inside Anthropic, although later than predicted. His comments also included a qualification: programmers would still need to specify what was being built, how it fit with other code and whether the design was secure.

Anthropic’s internal figures should not be treated as a measure of the whole software industry: the company develops Claude and is an unusually intensive user of its own tools. Other cited figures use different measures and dates. 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. Those figures describe distinct organizations and measures, not a single industry-wide trend.

Experiments also test different kinds of work. A 2022 study of 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 involved real tasks in large projects familiar to the participants. Different task sizes, codebases and tool generations help explain why results should not be collapsed into one claim about developer productivity.

“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

What the Evidence Cannot Yet Show

Industry-wide productivity remains unclear. Anthropic’s code share and output figures are company-reported measures from a firm that builds AI coding tools; they do not establish comparable gains across employers or prove that more merged lines mean more valuable software. The controlled studies differ in date, task and setting, while METR’s newer estimates are too uncertain to pin down a reliable speed effect.

The hiring figures also do not settle whether AI caused fewer entry-level opportunities. A lower share of junior postings and a decline in employment among young workers in AI-exposed jobs are findings reported in the source material, but neither alone establishes the cause or what happened to the people affected. The available account does not provide enough detail to assess how results vary by occupation, employer or individual skill level.

It is also not known whether AI-assisted learners will retain knowledge better or worse over longer periods, or how the observed quiz gap translates into work performance. The evidence supports caution about replacing practice with generated answers, but does not establish that using AI prevents people from learning to code.

Watch Hiring and Learning Data

The next useful signals will be comparable, independently measured studies of real software work as tools change, alongside longer-term data on code quality, defects and time saved. METR’s account describes its later results as weak evidence, so stronger trials would be needed before treating a precise productivity gain as established.

Hiring trends will also need continued monitoring. Indeed’s November 2026 posting increase followed a May 2025 low, but a single recovery measure does not show whether entry-level openings will rise or whether junior workers’ employment will improve. Future data may clarify whether employers are changing the number of roles, the skills they seek, or both.

For learners and managers, the evidence so far favors pairing AI use with hands-on practice and review. Whether that approach produces stronger long-term skills or better job outcomes remains to be tested. The immediate takeaway is narrower: AI can generate substantial amounts of code, while the value and limits of human programming knowledge remain an open, practical question.

Key Questions

Does AI make learning to code a waste of time?

No evidence cited here establishes that learning to code is a waste of time. AI can generate code, but people still need to specify requirements, review results, test behavior and debug failures. An Anthropic learning experiment found lower quiz scores among AI users than among developers who coded by hand, though it does not prove AI always harms learning.

Does AI make software developers faster?

The results are mixed. A 2025 METR randomized trial found experienced developers took 19% longer with AI on the tasks tested. A later study suggested possible speed gains, but its estimates were uncertain and included no effect. A separate 2022 experiment found faster completion on a small coding task, illustrating how results depend on the work being measured.

How much code does AI write at Anthropic?

Anthropic said Claude authored more than 80% of code merged into its codebase in May 2026, up from the low single digits before Claude Code launched in February 2025. This is a company-reported internal measure, not an estimate for the software industry as a whole.

Are entry-level software jobs disappearing?

The cited data show entry-level roles were 4.5% of US software development postings on Indeed in the first quarter of 2026, while senior roles were 69.3%. The figures indicate a difficult entry-level market but do not establish that AI caused the pattern or that junior jobs are disappearing.

Source: ThorstenMeyerAI.com

You May Also Like

GPT-6.2 Astra And Sol: What’s Changed In Cost, Scores, And Work?

OpenAI’s GPT-6.2 Astra and Sol score strongly in independent tests, but show mixed results, cybersecurity risks and different costs.

Apple Plans Camera AirPods Alongside Upgraded Foldable iPhone in 2027

Apple is expected to release a new foldable iPhone and camera-enabled AirPods in 2027, according to Bloomberg rumors. Details remain uncertain.

Anthropic’s Claude Designs Leaner Quantum Circuits, Reducing Resources Needed For Key Calculations

Anthropic and Harvard researchers report AI-designed quantum circuits that cut resource estimates across eight tasks, but none was tested on quantum hardware.

Unpacking Anthropic’s Contributions To AI And The Debates Surrounding It

Analysis of Anthropic’s contributions to AI and the controversies surrounding its practices and products, based on recent reference entries and industry context.