🔍 Read the full analysis: Coding In The Age Of AI: Questions For Students, Developers And Managers on ThorstenMeyerAI.com
Get audio and creator gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
AI now authors a large share of code at some companies, but evidence that it reliably makes developers faster is mixed. Hiring data point to a tight entry-level market, while an experiment found that developers who used AI to learn a new library scored lower on a follow-up quiz than those who coded by hand.
A review of 2026 job-market data, company disclosures and coding studies finds that AI can author much of the code developers merge, but there is not yet strong evidence that it consistently makes them faster. The findings matter to students, working developers and hiring managers weighing what skills to teach, use and recruit for as entry-level software postings remain a small share of listings.
Anthropic said that in May 2026, Claude authored more than 80% of code the company merged into its codebase, compared with a share in the low single digits before Claude Code 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. 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 from controlled tests is less conclusive. In a randomized METR trial involving 16 experienced open-source developers and 246 real tasks, developers using AI tools available from February to June 2025 took 19% longer to complete tasks, despite believing beforehand that AI would reduce their time. METR said a later study offered only “very weak evidence” about the size of any improvement by early 2026. In a separate Anthropic learning 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.
The labor-market picture is mixed. Indeed data cited in the report show that 4.5% of US software-development postings were entry-level in the first quarter of 2026, while 69.3% were senior. At the same time, US software-development postings were up about 23% year over year in November 2026, after reaching a low in May 2025. Stanford researchers found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it otherwise would have been, a finding that does not by itself establish AI as the sole cause.
Skills Beyond Code Generation
The findings suggest that writing code is no longer the whole job at workplaces adopting AI tools. Developers still need to describe what a system should do, understand how a change fits into an existing codebase, test outputs and identify defects or security risks. Those tasks become more consequential when a tool produces code faster than a person can carefully review it.
For students and career switchers, the figures do not show that learning programming is pointless. They do point to a more difficult entry route: junior listings are scarce in the reported data, and AI-assisted practice may not build understanding as reliably as hands-on work. For managers, code volume is not a sufficient measure of output. Quality, maintainability, security and the time required for review also matter.
These are implications of the reported evidence, not proof that every employer will change hiring or training in the same way. Anthropic’s internal figures describe a company that builds AI coding tools and uses them heavily, while the controlled studies cover specific tasks and groups of developers.
Products Worth Considering

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue

Sony WH-1000XM6 Wireless Noise-Canceling Headphones, Black
As an affiliate, we earn on qualifying purchases.
A Rapidly Changing Evidence Base
Anthropic chief executive Dario Amodei said in March 2025 that AI would write 90% of code in three to six months and essentially all code in twelve months. His forecast included a qualification: programmers would still need to specify what was being built, how it connected to other code and whether the design was secure. Anthropic’s later internal data indicate substantial use of Claude at the company, but do not establish that the broader industry has reached the same level.
Other published figures offer a different scale and come from different periods. In October 2024, Google chief executive Sundar Pichai said more than a quarter of new code at Google was generated by AI and then reviewed and accepted by engineers. METR cited an estimate in February 2026 that Claude Code authored approximately 4% of GitHub commits. These measures cover different companies, tools and definitions, so they cannot be treated as a direct comparison.
Results also depend on the task. A 2022 experiment with 95 professional developers found that participants using GitHub Copilot completed a small JavaScript HTTP-server task 55.8% faster on average. METR’s later trial involved real tasks in established projects familiar to experienced developers. A short, self-contained coding exercise may not capture the work of understanding and changing a large codebase.
“The programmer still needs to specify what’s 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
Limits of the Available Evidence
The industry-wide share of AI-authored code is unknown. Anthropic’s figures describe its own codebase, while Google’s statement and the GitHub estimate use different measures and dates. They do not show how much code AI writes across all software companies.
The productivity effect is also unresolved. METR said its newer study was an unreliable signal because some developers were reluctant to take tasks they did not want to do without AI. Its estimates included confidence intervals that crossed no effect. The available results therefore do not establish a dependable speed-up across developers, tools or kinds of work.
The employment data show weak entry-level representation and lower employment among younger workers in AI-exposed jobs, but the reported figures do not establish how much of either pattern was caused by AI rather than other labor-market factors. The material does not provide a comparable long-term measure of software-job demand or evidence that the learning-experiment result applies to every student or coding tool.
What Schools and Employers Track
The next useful evidence will come from larger, repeated studies that test newer AI tools on varied software tasks and measure completion time alongside correctness, security, maintenance and review effort. METR has said its early-2026 evidence cannot yet provide a reliable estimate of the productivity effect, leaving room for better-designed follow-up work.
Hiring data will also need continued monitoring. The November 2026 increase in US software-development postings followed a sharp decline from the May 2025 low, but one year-over-year figure cannot show whether entry-level opportunities are recovering. Schools and employers can compare hands-on and AI-assisted training by testing whether learners retain skills in debugging, code review and independent problem-solving.
For now, the evidence supports a measured approach: learn programming fundamentals, practice reading and testing generated code, and treat claims about productivity as specific to the company, task and measurement used. Whether AI expands or narrows junior career paths over time remains unsettled.
Key Questions
Does the evidence show that students should stop learning to code?
No. The report argues that coding remains useful, while the work is shifting toward specifying, testing, debugging and reviewing code as well as writing it. The available evidence does not show that programming education is no longer needed.
Does AI make developers faster?
Not consistently in the evidence described. METR’s early-2025 randomized trial found participants took 19% longer with AI on its tasks, while a separate 2022 experiment found faster completion on a small coding exercise. METR said its later data were too weak to pin down the newer effect.
What did the Anthropic coding figure measure?
Anthropic said Claude authored more than 80% of code merged into its own codebase in May 2026. This is a company-specific figure, not an estimate for the whole software industry.
What does the report say about entry-level software jobs?
Indeed data cited in the report show that entry-level roles made up 4.5% of US software-development postings in the first quarter of 2026. The figure describes that period and does not establish that AI alone caused the low share.
What remains uncertain about AI coding tools?
The size and consistency of productivity gains, the share of code AI writes across the industry and the long-term effect on junior hiring remain unclear. Studies vary by task, developer experience, tool and measurement method.
Source: ThorstenMeyerAI.com
