Is Coding Still A Useful Skill For Students And Managers?
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Is Coding Still A Useful Skill For Students And Managers? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

  • 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 is writing a large share of code at some companies, but research does not show that coding skills are no longer useful. Evidence instead points to the value of specifying, reviewing, testing and debugging code, while raising concerns about entry-level opportunities and whether AI makes developers faster on complex work.

AI is writing a growing share of software code, but available 2026 evidence does not show that learning to code has become pointless for students, developers or managers. Company figures, controlled trials and job-posting data instead suggest that the work is shifting toward specifying, reviewing, testing and debugging AI-generated code, while the effects on productivity and entry-level hiring remain unsettled.

Anthropic reported that in May 2026, Claude authored more than 80% of the code merged into the company’s codebase. The company said the share had been in the low single digits before its Claude Code tool launched in February 2025. It 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 this measure tracks code quantity rather than quality and is “almost certainly an overstatement” of the productivity gain.

Whether AI makes developers faster is less clear. In a randomized METR trial conducted in early 2025, 16 experienced developers completed real tasks in codebases they knew well. They took 19% longer with AI tools, despite expecting beforehand that AI would reduce their time and later believing it had made them faster. METR’s follow-up work suggested developers may have gained more from AI by early 2026, but the organization described its newer results as “only very weak evidence” about the size of any speed-up.

Labor-market indicators show pressure on early-career routes, but not an end to software hiring. Indeed data put entry-level roles at 4.5% of US software-development postings in the first quarter of 2026, compared with 69.3% classified as senior. The source also reports that US software-development postings were up about 23% year over year in November 2026, following a low in May 2025. Those figures describe postings, not hires, and do not establish that AI caused the change in job mix.

At a glance
analysisWhen: Evidence and figures reported through N…
The developmentA review of 2026 company data, labor-market indicators and experiments finds that AI is changing coding work, but the evidence does not support treating programming skills as obsolete.

The Skills Behind AI-Generated Code

For students and career switchers, the evidence argues against replacing programming education with tool familiarity alone. Knowing how code works helps a person identify when generated code does not meet requirements, contains a defect or creates a security risk. The central skill may increasingly be judging and directing code, rather than typing every line.

That judgment matters for managers as well as developers. A team may produce more code without delivering an equivalent increase in reliable, useful software. Managers deciding whom to hire or train need to distinguish output volume from quality, and to assess whether employees can explain, test and maintain the systems they build. Anthropic itself warns that lines of code are a limited productivity measure.

There is also a learning risk. In an Anthropic experiment, mostly junior developers using AI while learning a new library scored 50% on a follow-up quiz, compared with 67% among those who coded by hand. The reported largest gap was in debugging. The result does not prove that all AI-assisted learning is less effective, but it suggests that students who delegate the reasoning may have less opportunity to develop skills they will need when tools fail.

De'Longhi Magnifica Start Automatic Espresso Machine with Milk Frother, Silver

De'Longhi Magnifica Start Automatic Espresso Machine with Milk Frother, Silver

TCL F8D Series 75-Inch 4K QLED Smart Fire TV, 144Hz, HDR (75F8D, 2026)

TCL F8D Series 75-Inch 4K QLED Smart Fire TV, 144Hz, HDR (75F8D, 2026)

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. He also said programmers would still need to specify what to build, how it fits with existing code and whether the design is secure. The company’s later internal figures show substantial AI involvement at Anthropic, but do not establish that the same share applies across the industry.

Other published estimates offer a narrower view. 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 approximately 4% of GitHub commits. These numbers cover different organizations, periods and measures, so they are not directly comparable.

Experiments have also produced different results. A 2022 test involving 95 professional developers found that participants using GitHub Copilot completed a small JavaScript task 55.8% faster on average. METR’s later trial examined real tasks in established open-source projects, where familiarity, existing design choices and task complexity could affect results. The contrast shows why a result from a short coding exercise cannot automatically predict performance on ongoing software work.

“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 Productivity Data Cannot Show

The evidence does not settle how much AI changes productivity across different teams or tasks. Anthropic’s figures describe a company that builds and uses its own AI coding tools; its eightfold code-volume measure is not a verified eightfold rise in useful output. METR’s later trial produced estimates with confidence intervals that included no effect, and the organization said its results were weak evidence about the size of any improvement.

It is also unclear how much AI explains the narrow share of entry-level postings. The reported job data show a distribution and a change over time, while Stanford researchers’ estimate that employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it otherwise would have been is an analysis of exposure, not proof that AI alone caused the difference. The available figures do not establish what share of affected roles were software jobs or how many postings became actual hires.

Finally, the learning experiment covered a particular group and task: mostly junior developers studying a new library. It does not determine the effects of every teaching method or tool. More evidence is needed on whether structured AI use—such as asking learners to explain, test and repair suggestions—can preserve learning while providing assistance.

What Students and Managers Should Track

The next useful evidence will come from studies that measure completed, maintainable software rather than code volume alone, and compare results across task types, experience levels and AI tools. METR has said its later results are difficult to interpret as developer dependence on AI has grown; further controlled research could clarify whether speed gains are real and how large they are.

Students choosing what to learn can treat coding as a way to build understanding, not only as preparation to type code manually. Practice in reading unfamiliar code, checking outputs, testing edge cases, debugging and explaining design choices addresses the work that remains when AI generates a first draft. The current evidence supports those skills, but does not establish one ideal curriculum.

Managers can track quality, security, maintenance and delivery outcomes alongside output measures, and avoid assuming that more generated code means proportionally more productivity. Hiring and training decisions will also need to account for limited entry-level openings and the possibility that junior staff need deliberate opportunities to learn through hands-on work. Further job-posting and employment data will show whether the recent rise in software-development listings continues and whether entry-level access improves.

Key Questions

Is coding still worth learning?

The evidence reviewed here does not show that coding has become obsolete. It points to continuing value in understanding code so people can specify requirements, review outputs, test behavior and debug problems.

Does AI make software developers faster?

Results vary by task and study. METR’s early-2025 trial found experienced developers took 19% longer with AI on real tasks, while a 2022 Copilot experiment found faster completion on a short JavaScript exercise. METR said its later evidence for increased speed by early 2026 was weak.

Are entry-level software jobs disappearing?

Indeed data cited in the report put entry-level roles at 4.5% of US software-development postings in the first quarter of 2026. That indicates a small share of postings, but does not by itself prove that AI caused the pattern or that entry-level jobs have disappeared.

What should managers assess when teams use coding AI?

Alongside code volume, managers can assess whether software meets requirements, passes tests, is secure and can be maintained. Anthropic cautions that lines of code measure quantity rather than quality, so output counts alone can overstate productivity gains.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Fable 5 Is Back. GPT-5.6 Is Next. And Anthropic Reportedly Already Has Something Stronger.

Anthropic restores Fable 5 after government blackout; OpenAI previews GPT-5.6, and rumors suggest a more capable model already exists.

Apple Silicon’s Quiet Memory Advantage

Apple Silicon chips offer a unique memory architecture that enables larger AI models locally, with trade-offs in speed and bandwidth. Here’s what is confirmed.

How 14 AI Coding Assistants Can Fit Into Learning And Software Development In 2027

A comparison of 14 books on AI coding assistants, from beginner guides and multi-tool coverage to team workflows and specialist references.

How T3 Code Fits Claude, Codex, And Several Macs Into My Workflow

Thorsten Meyer says he now uses T3 Code to manage Claude Code and Codex across eight Macs and a Linux server, while noting beta features and limits.