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

🔍 Read the full analysis: How Students Can Decide Whether To Learn Coding As AI Grows 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 now writes a large share of some companies’ code, but research does not show that coding skills are no longer useful. Students face a tighter entry-level job market and should learn to read, test, debug and evaluate code as well as use AI tools.

A new review of 2026 job-posting data, workplace disclosures and coding studies finds that AI is writing a growing share of software code, but the evidence does not establish that learning to code is a waste of time. For students, the practical choice is less about whether to learn coding than how to combine it with specifying, reviewing, testing and debugging software.

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 its coding tool 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. Those are company-reported figures, not an independent measure of software quality or productivity. Anthropic cautioned that lines of code count quantity rather than quality and said the eightfold figure was “almost certainly an overstatement” of the productivity gain.

Independent evidence on speed is mixed. In a randomized METR trial conducted in early 2025, 16 experienced open-source developers completed 246 tasks in projects they already knew. Developers using AI took 19% longer, despite believing it had made them faster. METR said a later study offered only “very weak evidence” about the size of any speed-up by early 2026. A separate 2022 test of 95 professional developers found that participants using GitHub Copilot completed a small JavaScript task 55.8% faster, illustrating how results can depend on the task and setting.

The hiring picture is also uneven. Indeed data cited in the report show that 4.5% of U.S. software-development postings were entry-level in the first quarter of 2026, while 69.3% were senior. Stanford researchers, meanwhile, estimated that employment among 22- to 25-year-olds in jobs exposed to AI was 19% below where it would otherwise have been. These findings indicate pressure on early-career opportunities, but do not show that AI alone caused the changes. Indeed postings for U.S. software-development roles were up about 23% year over year in November 2026, after reaching a low in May 2025.

At a glance
reportWhen: Published November 18, 2026; the eviden…
The developmentA November 2026 report brings together company disclosures, job postings and research to examine whether students should learn coding as AI takes on more software development work.

Coding Skills Beyond Code Generation

For students choosing courses or career paths, the evidence points away from treating coding as a choice between manual programming and AI. If tools generate more code, a developer’s contribution can shift toward explaining what a system should do, fitting new code into an existing project and identifying errors or security risks. Those responsibilities require enough programming knowledge to judge what a tool produces.

There are risks in relying on AI before building that knowledge. In an Anthropic experiment, mostly junior developers learning a new library with AI scored 50% on a follow-up quiz, compared with 67% among participants who coded by hand. The largest gap concerned debugging. The result comes from one experiment and should not be treated as a universal measure of AI-assisted learning, but it suggests that getting working code quickly may not mean understanding it well enough to maintain or troubleshoot it.

Students and career switchers also need to account for a more difficult entry-level market. Fewer junior postings can make it harder to get a first role, even as overall software-development postings rise. Building demonstrable skills in testing, debugging and explaining design decisions may help show what a candidate can do, though the cited evidence does not establish which skills will guarantee employment.

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

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

Ninja Mega Kitchen System Blender and Food Processor, 1500W, 72 oz Pitcher, 8-Cup Bowl, Black (BL770)

Ninja Mega Kitchen System Blender and Food Processor, 1500W, 72 oz Pitcher, 8-Cup Bowl, Black (BL770)

As an affiliate, we earn on qualifying purchases.

What the Studies Actually Measure

Anthropic chief executive Dario Amodei said at the Council on Foreign Relations on March 10, 2025, that AI would be “writing 90 percent of the code” in three to six months and “essentially all of the code” in twelve. His forecast is a claim, not a verified industry-wide measure. The later Anthropic disclosure offers evidence about one company’s own codebase, but Anthropic builds Claude and may use its tools more heavily than other employers.

Other reported figures cover different populations and measures. Google chief executive Sundar Pichai said in October 2024 that AI generated more than a quarter of Google’s new code, which engineers then reviewed and accepted. METR cited an estimate in February 2026 that Claude Code authored about 4% of GitHub commits. Company code shares, repository commits, task-completion times and job postings do not measure the same thing, so they cannot be combined into a single estimate of AI’s effect on software work.

The productivity studies differ in their tasks, too. The 2022 Copilot test asked developers to build a small HTTP server, while METR’s trial involved real tasks in large, familiar open-source projects. A tool can speed up a bounded exercise without necessarily speeding up work that involves understanding a complex codebase, checking behavior or meeting project 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 on March 10, 2025

How Jobs and Productivity May Shift

It remains unclear how quickly AI will change the number and type of software jobs, or whether the decline in entry-level postings is primarily driven by AI. The Stanford estimate concerns young workers in AI-exposed jobs and does not, by itself, isolate AI as the cause. A rise in software-development postings by November 2026 also does not show how many jobs are newly created, junior-level or suitable for new graduates.

Productivity estimates are unsettled as well. METR’s early-2025 trial found slower completion with AI, while its later data offered weak evidence of possible gains. Anthropic’s internal code-volume figures come from one AI company and do not establish that other teams will see similar results. The available material does not settle which programming subjects, learning methods or credentials will best prepare students for future hiring.

What Students Can Track Next

Students can follow whether future labor-market data show a sustained recovery in entry-level software postings, rather than relying on overall job counts alone. New controlled studies and independently measured workplace results may also clarify when AI saves time and whether developers retain what they learn while using it.

For now, a practical course of study can pair programming fundamentals with regular use of AI tools while requiring students to explain, test and debug the resulting code. Learners should also practice defining requirements, checking how a change affects an existing system and identifying security concerns. These are not guarantees of a job; they are ways to build judgment alongside tool familiarity as the evidence develops.

Key Questions

Is learning to code still worthwhile?

The cited evidence does not show that coding has become useless. AI can generate code, but people still need skills to specify requirements, review results, test behavior and diagnose problems.

Does AI consistently make developers faster?

No consistent result is established. METR’s early-2025 trial found developers took 19% longer with AI, while a 2022 test found faster completion on a small coding task. Results depend on the task, and METR described its later evidence on speed-ups as weak.

Are entry-level software jobs disappearing?

The report cites a low share of entry-level U.S. software-development postings in the first quarter of 2026, alongside a later rise in overall postings. Those figures show a challenging and changing market, but they do not prove that entry-level jobs are disappearing or that AI caused the decline.

What should students focus on learning?

Alongside programming fundamentals, students can practice reading and reviewing code, testing, debugging, defining what software should do and evaluating design and security. The cited evidence does not establish a single curriculum that guarantees employment.

Source: ThorstenMeyerAI.com

You May Also Like

Blender’s UI And Runtime: Much Faster Than GPUI

A GitHub project packages Blender-derived widgets and interactions as a standalone C runtime, with language bindings and developer-published performance claims.

We Halved DNS Build Delay And Cut Memory Use By 85%

Cloudflare says targeted DNS builds cut peak memory use by 85% and halved the wait before changes are written to Quicksilver.

The Tilde In Your PATH May Not Be Your HOME

A shell sandbox warning highlighted that quoted ~/.local/bin in PATH can stay a literal tilde. Use $HOME and check your PATH entries.

How to Choose Software Testing Tools

Set up a practical software testing toolkit, run checks on your project, and confirm the results before release.