🔍 Read the full analysis: Learning To Code In The AI Era: A Guide For Students And Professionals on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Evidence available through November 2026 shows that AI can author a large share of code at some companies, but that does not establish equal gains in productivity or mean coding skills are no longer needed. Entry-level software job postings remain a concern, while experiments suggest AI-assisted learning can leave gaps in understanding.
A review of 2026 job-posting data, developer experiments and company disclosures finds that AI is writing a substantial share of code at some technology companies, but the evidence does not show that learning to program has become unnecessary. The findings matter to students, career changers and employers because they point to changing expectations for junior developers, while leaving the size of AI’s productivity gains uncertain.
Anthropic reported that Claude authored more than 80% of code merged into the company’s codebase in May 2026, compared with what it described as 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 eight-fold figure is “almost certainly an overstatement” of the productivity gain.
Independent evidence on speed is less conclusive. In a randomized METR trial conducted in early 2025, 16 experienced open-source developers completed 246 tasks in familiar projects. They took 19% longer when allowed to use AI tools, even though participants believed the tools had made them faster. METR said in February 2026 that a later study suggested developers may be getting more help from AI, but described the newer results as “only very weak evidence” about the size of any improvement.
The labor picture is mixed. 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. Yet U.S. software-development postings were up about 23% year over year in November 2026, after reaching a low in May 2025. These figures describe postings, not hires, and do not establish that AI caused changes in demand.
Skills Beyond Code Generation
The findings suggest that coding education may need to put more weight on work that surrounds code generation: defining requirements, reading code, testing changes and debugging failures. These skills help a developer check whether an AI-generated answer fits an existing system and behaves as intended. They also matter when a security-sensitive design must be reviewed rather than accepted because it appears plausible.
For students and career switchers, a thinner share of entry-level postings may make it harder to get a first role, although the available figures do not prove that AI is the cause. For experienced developers, AI may change how work is performed without removing the need for technical judgment. Employers deciding how to train or hire staff face the same distinction: code volume alone is not a reliable measure of useful output.
An Anthropic experiment adds a caution for education. 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 result comes from a particular experiment, not a universal measure of learning, but it suggests that producing working code with AI and understanding the underlying concepts are not necessarily the same thing.
Popular right now

Dyson V9 Motorbar Cordless Vacuum

Ninja Prestige Smart XL 10-in-1 Air Fryer and Countertop Oven with Pro Cook System, Stainless Steel, DT551

Samsung 65-Inch 4K OLED Smart TV with Dolby Atmos, 120Hz Refresh Rate, and Amber Protection Bundle
As an affiliate, we earn on qualifying purchases.
Code Generation Versus Productivity
Anthropic chief executive Dario Amodei said at the Council on Foreign Relations on March 10, 2025, that AI would be writing 90% of code within three to six months and essentially all code within 12 months. The company’s later figures indicate extensive use of Claude inside Anthropic, but they measure code authorship at one employer, not the share of software work across the industry.
Other published figures offer different snapshots. 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 about 4% of GitHub commits were authored by Claude Code. The measures and dates differ, so they should not be treated as a direct comparison of industry-wide adoption.
Experiments also depend on the task. A 2022 study of 95 professional developers found that participants using GitHub Copilot finished a small JavaScript HTTP-server task 55.8% faster on average. METR’s later trial involved real tasks in large, familiar codebases. The contrast shows why results from a short, self-contained exercise may not predict the time required for work inside complex software projects.
“More than 80% of the code we merge into Anthropic’s codebase was authored by Claude.”
— Anthropic, in its report on code production
Limits of the Available Evidence
Anthropic’s code-production statistics come from a company that builds and heavily uses AI coding tools. They do not establish how common the same level of use is elsewhere, or how much of the additional code improves products. Its output figures are company estimates, and the company itself warns against treating lines of code as a direct measure of productivity.
METR’s early-2026 results are also uncertain. In its February report, the group said some developers increasingly declined tasks they did not want to do without AI, complicating the study. Its raw estimates pointed to tasks taking 18% less time for developers from the earlier study and 4% less for newly recruited participants, but the confidence intervals included no effect. The report therefore does not settle whether AI now speeds up typical software work, or by how much.
The labor figures do not answer whether AI is responsible for a smaller share of entry-level postings, and postings are not the same as completed hires. The available learning experiment also covers a particular group and task; its result cannot establish how all students learn with AI. Longer-term effects on hiring, skill development and software quality remain unclear.
Track Hiring and Learning Results
The next useful evidence will include comparable, independent studies that test AI tools on real software work and measure quality as well as completion time. Researchers and employers will also need to distinguish code generated from work successfully reviewed, integrated and maintained.
Students, professionals and managers can watch for updated job-posting and hiring data, along with studies that follow whether AI-assisted learners retain concepts and can debug unfamiliar code. Until those results are available, the evidence supports neither abandoning programming education nor assuming that AI use automatically makes developers faster. The current picture is a changing division of work, with the size and long-term effects still developing.
Key Questions
Does AI mean students should stop learning to code?
No. The evidence in the report points to continuing value in understanding, testing and debugging code, even when AI helps produce it. It does not establish that coding knowledge is no longer needed.
Does AI make professional developers faster?
The results are mixed. METR’s early-2025 randomized trial found participants took 19% longer with AI, while the group’s later data suggested possible time savings but were too weak to establish their size.
Are entry-level software jobs disappearing?
The cited Indeed data show entry-level roles made up 4.5% of U.S. software-development postings in the first quarter of 2026. That is a posting share, not a count of hires, and it does not show that AI caused the pattern.
What should learners practice alongside AI tools?
The findings support practicing how to specify a task, read and review generated code, test behavior, debug errors and judge design choices, including security. These skills help learners assess whether a generated solution is correct and appropriate.
Source: ThorstenMeyerAI.com
