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

🔍 Read the full analysis: How Managers Can Evaluate The Need For Coding Skills On Their Teams on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

AI is writing a growing share of code at some companies, but available evidence does not show that coding judgment is no longer needed. Managers should define coding needs by role and assess whether staff can specify, review, test and debug AI-generated work, while recognizing that productivity gains and effects on hiring remain uncertain.

A November 18 analysis by ThorstenMeyerAI.com argues that managers should assess what coding work their teams need to do, rather than assume AI tools make programming skills unnecessary. The case rests on mixed evidence: Anthropic says Claude authored more than 80% of code merged at the company in May 2026, while studies of developer speed and learning show that code production alone does not establish whether work is faster or sound.

For managers, the analysis recommends looking beyond whether employees can type code. It points to skills in specifying requirements, understanding existing systems, reviewing output, testing and debugging as important when AI produces part of the implementation. These abilities help a team decide whether generated code meets the actual need and fits securely into a larger system.

The evidence on speed is mixed. A randomized METR trial involving 16 experienced open-source developers and 246 tasks found that AI tools increased completion time by 19% in the early-2025 study. Developers had expected a 24% time reduction and afterward believed they had saved 20%. METR reported in February 2026 that a later study suggested possible time savings, but its estimates were uncertain and its evidence for the size of any improvement was weak.

Learning is another management concern. In an Anthropic experiment, mostly junior developers using AI to learn a new library scored 50% on a follow-up quiz, compared with 67% among developers who coded by hand; the largest gap concerned debugging. The analysis says managers should consider whether AI-assisted workflows are helping employees build the knowledge needed to maintain and evaluate code, not just complete immediate tasks.

At a glance
analysisWhen: Published November 18, 2026; the cited…
The developmentA November 18, 2026, analysis from ThorstenMeyerAI.com examines research and workplace data to explain how managers should evaluate coding skills as AI tools write more software.

Hiring for Judgment, Not Keystrokes

The decision affects hiring criteria, training plans and how managers measure productivity. If teams assess only how quickly someone produces code, they may overlook whether that person can catch incorrect assumptions, security problems or failures in unfamiliar parts of a system. Conversely, requiring every role to involve extensive manual coding could screen out people whose work depends more on architecture, product requirements or review.

Labor-market figures add pressure but do not settle the question. Indeed data cited in the analysis show that 4.5% of US software-development postings were entry-level in the first quarter of 2026, while 69.3% were senior. The analysis also reports Stanford researchers’ finding that employment among 22- to 25-year-olds in AI-exposed jobs was 19% below where it would have been. Those figures describe a difficult entry-level market, but they do not establish that AI alone caused the change or that coding skills are no longer useful.

For managers, a practical implication is to tie skill requirements to actual tasks. Teams can distinguish roles that need people to build and modify software directly from those where employees mainly configure tools, define requirements or check results. The available evidence supports making that distinction; it does not provide a universal hiring formula.

Products Worth Considering

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

As an affiliate, we earn on qualifying purchases.

AI Code Shares and Their Limits

The strongest code-share figure in the analysis comes from Anthropic’s own reporting. The company said Claude authored more than 80% of code it merged in May 2026, compared with a figure in the low single digits before Claude Code launched in February 2025. Anthropic also reported that the 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 productivity gains. The company builds Claude and is not necessarily representative of other employers. Other figures cited are from different settings and dates: Google CEO Sundar Pichai said in October 2024 that AI generated more than a quarter of new Google code, which engineers then reviewed and accepted. These measures cannot be treated as a direct comparison across companies.

Study design also matters. A 2022 experiment with 95 professional developers found that participants using Copilot completed a small JavaScript HTTP-server task 55.8% faster. That result differs from METR’s trial of experienced developers working on real tasks in codebases they already knew. A short, self-contained task and work inside a mature project test different abilities, so neither result alone describes every team’s likely experience.

“The code is written by Claude, with the engineer directing and reviewing, rather than typing it themselves.”

— Anthropic, in its report on code production

Productivity and Hiring Remain Unsettled

The evidence does not establish a single productivity effect that applies across teams. METR’s controlled trial found slower completion with AI in early 2025, while its later estimates had confidence intervals that included no effect. The organization also said developers increasingly resisted doing tasks without AI, complicating its later study. The size of any change as tools improved remains uncertain.

Anthropic’s code-share and output figures are company-reported measures from a firm that develops AI coding tools. They do not show that more code is better code, or that the same results would occur at a typical employer. The entry-level job figures and employment estimates also describe trends, not proof that AI caused them. The supplied material does not establish which specific coding skills every manager should require for each role.

Match Skills to Team Tasks

Managers can apply the analysis by mapping each role to its expected work: writing or changing code, defining requirements, integrating systems, reviewing generated output, testing, or debugging. They can then set hiring and training expectations around those tasks and assess whether employees can explain and verify the work they deliver. That is a practical approach, not a benchmark validated by the cited studies.

The next evidence to watch is whether larger, independent studies measure AI-assisted work across different experience levels, software projects and task types. Updated labor-market data may also clarify whether the entry-level pattern persists. Until then, teams should treat AI code-share figures as evidence of adoption, not as proof that programming expertise or review can be removed.

Key Questions

Does the evidence show that teams no longer need coding skills?

No. The analysis says AI can write a substantial share of code in some settings, but managers still need to evaluate requirements, code quality, security, testing and debugging. The evidence does not support dropping coding judgment as a general requirement.

Did AI make developers faster in the METR trial?

In METR’s early-2025 randomized trial, developers took 19% longer to complete tasks with AI tools, even though they believed the tools had made them faster. METR’s later evidence pointed toward possible speed gains but was too weak to establish their size.

What should managers assess when hiring or training?

Requirements depend on the role, but the analysis highlights the ability to specify what should be built, understand surrounding code, review AI output, test and debug. Managers can map those capabilities to the work each position actually involves.

Does Anthropic’s 80% figure represent the whole software industry?

No. It is Anthropic’s company-reported figure for code merged into its own codebase in May 2026. The analysis cautions that Anthropic builds AI tools and that its results may not represent other employers.

Source: ThorstenMeyerAI.com

You May Also Like

Palo Alto Networks Targets AI Computing With Applied Compute Acquisition

Palo Alto Networks agreed to acquire Applied Compute in an all-stock deal valued at $4.5 billion, expanding into AI model training and deployment.

OpenAI’s Chromatic Codex Edition Takes Prompts From Screen To Cartridge

OpenAI’s $200 ModRetro Chromatic: Codex Edition went on sale Nov. 9. The handheld includes a programmable cartridge for Codex-built games.

A Closer Look At DaisyRose-27B’s AI Post-Training Methods

RadixArk says post-training DaisyRose-27B raised its Terminal-Bench 4.1 pass@1 score to 12.0%; independent checks and evaluation details remain outstanding.

What’s Behind Grok’s Faster Limits For Users?

A headline says Elon Musk explained faster Grok usage-limit hits, but the explanation, affected users and any policy change remain unverified.