🔍 Read the full analysis: Do Managers Need Coding Skills In The Age Of AI? on ThorstenMeyerAI.com
Get smart everyday buys 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 writes a large share of code at some companies, but evidence that it consistently makes developers faster remains mixed. For managers, the emerging need is not necessarily to write every line, but to understand enough code to specify work, evaluate results and spot security or quality risks.
Managers overseeing software work still need coding fluency in the age of AI, according to evidence reviewed in a November 18 report by ThorstenMeyerAI.com: AI can generate substantial amounts of code, but people remain responsible for defining requirements, reviewing output and judging design risks. The findings do not establish that every manager must be a hands-on programmer, and studies of whether AI reliably speeds up development remain mixed.
The strongest public example of AI-generated code comes from Anthropic, which said that more than 80% of the code merged into its codebase in May 2026 was authored by Claude. The company said the share had been in the low single digits before Claude Code 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.
Those figures describe Anthropic’s own operations, not the software industry as a whole. The company cautioned that lines of code measure quantity rather than quality and that its eightfold increase in code per engineer almost certainly overstates the productivity gain. In a March 2026 employee poll, the median estimate was about four times as much output with AI; Anthropic said the actual uplift was likely lower. Google CEO Sundar Pichai said in October 2024 that AI generated more than a quarter of new code at Google, which engineers reviewed and accepted.
Controlled tests also complicate claims that more generated code means faster work. In a METR randomized trial, 16 experienced open-source developers completed 246 tasks in projects they already knew. With AI tools available, they took 19% longer, despite believing the tools had made them 20% faster. METR’s later study produced estimates pointing to time savings, but its confidence intervals included no effect, and the group described the results as weak evidence about the size of any improvement.
The Skills Managers Still Need
For managers, the practical issue is less whether they personally type code and more whether they can set precise requirements and assess what AI produces. A manager who cannot follow the logic of a proposed change may struggle to distinguish a working feature from code that merely appears plausible, or to ask whether it fits existing systems and security expectations.
The distinction matters for hiring and training, too. AI may let teams produce code with fewer manual steps, but it does not remove accountability for decisions about reliability, testing, maintenance or risk. The source report points to reading, testing and debugging code as skills with continued value, alongside the ability to describe what a system should do. That does not prove all managers need professional-level programming ability; it supports treating technical literacy as a tool for oversight rather than assuming AI makes it unnecessary.
There is also a learning risk. 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% for those who coded by hand. The largest difference was in debugging. The result suggests that managers planning AI-assisted training may need to protect time for people to understand and troubleshoot the code, rather than judging learning only by how quickly a task is completed.
Popular right now

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Dyson Airwrap i.d. Multi-Styler with Bluetooth Connectivity

DREAME D20 Air Plus Robot Vacuum and Mop with 20,000 Pa Suction
As an affiliate, we earn on qualifying purchases.
Hiring Signals And Earlier Tests
The employment picture is mixed rather than a simple story of demand disappearing. Indeed data cited in the report showed that 4.5% of US software-development job postings were entry-level in the first quarter of 2026, while 69.3% were senior. Stanford researchers found employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it otherwise would have been. Those findings describe specific measures and groups; they do not by themselves establish that AI caused every change.
At the same time, Indeed’s US software-development postings were about 23% higher year over year in November 2026, after reaching a low in May 2025. The source report says this points to continued demand, while a narrower entry-level share may make the first step into the field harder. The figures do not settle how much of either movement reflects AI rather than broader hiring conditions.
Research results have varied by task. A 2022 test involving 95 professional developers found those using GitHub Copilot completed a small JavaScript HTTP-server task 55.8% faster than those without it. METR’s later study involved work in large, familiar codebases, where developers had to account for project-specific rules. The contrast cautions against treating results from a short, self-contained coding exercise as a forecast for complex work in production systems.
“Almost certainly an overstatement of the true productivity gain.”
— Anthropic, on its reported increase in code per engineer
What The Evidence Cannot Yet Settle
How much AI improves real-world productivity remains unsettled. Anthropic’s internal output figures are self-reported company data and may not represent other employers. METR’s early-2025 trial found slower completion with AI, while its later study was too uncertain to give a firm estimate of the effect. The tasks, developers and tools differed across studies, making direct comparisons difficult.
It is also unclear how much of the recent employment pattern is attributable to AI. The entry-level posting share, Stanford’s employment estimate and Indeed’s year-over-year posting change measure different things. The source material does not establish a causal link between AI and every change in hiring, nor does it show exactly which coding abilities managers will be expected to have across industries.
The learning experiment involved a particular new-library task and mostly junior developers. It raises a concern about relying on AI during instruction, especially for debugging, but does not show that all AI-assisted learning produces weaker results. More evidence across different tasks, experience levels and workplaces is needed.
What Managers Should Watch
The next useful evidence will come from studies that track completed work, quality, defects and maintenance costs, not just lines of generated code or developers’ impressions of speed. Employers’ hiring patterns will also show whether entry-level roles recover, continue to shrink or change toward different expectations.
For now, managers can distinguish between coding fluency and being a full-time developer. The evidence supports learning enough to discuss requirements, inspect changes, ask about testing and security, and recognize when output needs deeper review. It does not establish a single required level of coding skill for every management role. Decisions about training and hiring will depend on the work being overseen and on whether AI-generated results can be verified.
Key Questions
Do managers need to know how to code?
The evidence does not show that every manager needs professional programming skills. It does support enough coding fluency to specify work, understand reviews and question risks when AI generates code.
Does AI make software developers faster?
Results are mixed. METR’s early-2025 trial found experienced developers took 19% longer with AI on its tasks, while a 2022 Copilot test found developers finished a small coding task 55.8% faster. Differences in tasks and tools limit direct comparison.
Is AI already writing most code?
Anthropic reported that Claude authored more than 80% of code merged into its own codebase in May 2026. That is a company-specific figure, not a measure of software development across the industry.
Should students still learn to code?
The report’s evidence points to continued value in learning how code works, including reading, testing and debugging. An Anthropic experiment found weaker quiz results among mostly junior developers who used AI while learning a new library, particularly on debugging.
Source: ThorstenMeyerAI.com
