👋 Introduction

Hello everyone. It seems that we cannot go one week without Cloudflare having an incident or a malicious actor adding a back door to a popular NPM package.

But first, a recent Stanford study has found out that AI is already impacting jobs, especially for college graduates:

🔍 Deep Dive: Employment Effects of Artificial Intelligence

“Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence” study available here

Stanford researchers Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen have released groundbreaking findings about artificial intelligence's impact on American employment, analyzing data from millions of workers through July 2025. Their research provides the first large-scale, real-time evidence of AI's employment effects using administrative payroll records from ADP, the largest payroll software provider in the United States.

Key Findings

The Stanford study focuses on changes in the labor market, particularly how widespread adoption of generative AI has impacted jobs for early career workers (ages 22 - 25), causing a 13 percent relative decline in employment.

This contrasts sharply with more experienced workers in the same occupations, whose employment has remained stable or continued growing. Software developers aged 22-25 saw employment decline by nearly 20% from late 2022 to July 2025, while older workers in the same field experienced growth.

The research identifies a critical distinction between AI applications: employment declines occur primarily in occupations where AI automates tasks rather than augments human work. Using data from Anthropic's Claude AI model, researchers found that occupations with high automation potential have shown significant job losses for young workers, while those with augmentative AI uses actually experienced employment growth.

Methodology and Robustness

The researchers analyzed monthly payroll records for 3.5 to 5 million workers, linking employment data to established AI exposure measures from multiple sources. They employed sophisticated statistical controls to rule out alternative explanations such as industry-specific shocks or interest rate changes. The findings remained robust across various sample constructions, excluding technology firms, remote-work occupations, and different demographic groups.

Broader Implications

While overall employment continues growing robustly, the stagnation in young worker employment growth since late 2022 appears driven by declines in AI-exposed positions. The researchers observe that labor market adjustments occur primarily through employment changes rather than wage adjustments, suggesting potential wage stickiness in the short term.

The study's "canaries in the coal mine" metaphor suggests these early effects on specific occupations may herald broader labor market transformations. The researchers theorize that AI primarily replaces codified knowledge—the "book learning" central to formal education—while being less capable of replacing the tacit knowledge accumulated through experience. This explains why younger workers, who rely more heavily on recently acquired formal skills, face greater vulnerability to AI displacement.

🚀 New Developments: Javascript ecosystem supply chain attacks

Multiple attacks happened lately, thankfully with minimal impact

The first attack happened on 8th September and impacted 18 NPM packages with 2 BILLION weekly downloads. These include chalk, strip ansi and debug, not random libraries. These packages are in pretty much everything.

The malware was a crypto clipper built to steal funds, which hijacked destination addresses of transactions, replacing them with an attacker’s wallet. Thankfully they "only" stole about $20 in crypto. But this shows that maybe it wasn’t an attack but instead a proof of concept.

As I highlighted in my X thread, the problem might not be on the packages themselves, but the fact that Node apps have thousands of dependencies and no one knows what 80% of them do.

Then on 16th September, the @​ctrl/tinycolor package was compromised to include a trojan malware, which affected 40+ other packages. Then another wave of supply chain attack hit targeting Crowdstrike packages.

What can you do to safeguard against these attacks?

  1. NEVER gitignore lock files

    1. package-lock.json

    2. yarn.lock

    3. pnpm-lock.yaml

  2. Ban wildcard versions in package.json. One allows any version. The other allows only patches.

    "chalk": "*"

    "lodash": "4.x"

    "chalk": "^4.1.2"

  3. Add npm audit to CI/CD. But your pipeline will fail immediately. You'll have 50+ vulnerabilities and half can't be fixed without breaking changes. This is the technical debt no one talks about.

🔥 Hot Topics: Cloudflare on fire, literally

Tough past couple of weeks for the infrastructure giant + other companies mishaps

First, there was an issue with unauthorized issuance of TLS certificates for their 1.1.1.1 IP address.

Then they ended up DDoSing themselves because of a React useEffect hook bug.

Hopefully they will sort this out soon since people are starting to have doubts in their ecosystem. It also doesn’t help that they have a monopoly on CDNs, with a market share of over 16% of all websites globally using the service, and with a stagering 82.16% share of the DDoS and bot protection market.

BONUS: Ubuntu Rust reimplementation goes poorly

Ubuntu, the most popular Linux distro, wanted to replace some of their code base from C/C++ with Rust. Turns out some code does not work as expected.

The Rust version of cksum is up to 17x slower than the GNU version for some large files. And the sort utility does not finish for large one line files. Turns out that Rust isn’t great for everything.

BONUS 2: Claude was indeed getting dumber

Anthropic has identified three issues from August to early September that affected the performance of it’s models. Turns out AI can get dumb.

Luckily, a Redditor created a this AI Benchmark tool scans models every hour to check if they are getting “dumber”. It is open source too!

📈 Recent Trend: MCP servers getting traction

Nuxt UI 4, the intuitive Vue UI library, has recently added an MCP server which will help AI tools like Cursor and Claude Code better understand the library and create user interfaces more easily.

Laravel, the popular PHP framework, also introduced Laravel Boost, which is an MCP server that runs in your project and that provides AI agents with Laravel-specific tools, documentation, and guidelines, enabling them to behave like experienced Laravel developers rather than generic search engines.

Not to be left behind, Symfony also released their MCP SDK in collaboration with the PHP Foundation and Anthropic. They also previously released the AI Bundle, which integrates various Symfony AI components such as the Symfony AI Platform component.

🔄 Tech Updates

🗝 Legacy Revival

  • IPv4 was supposed to be depleted 15 years ago, yet IPv6 is still not widely used

  • PHP 8.5.0 Beta 3 was released, while 83% of PHP websites still use legacy versions

  • Django release version 6.0 alpha 1, the first preview of the next major version

  • Expo released SDK verion 54 of the full-stack React Native framework

  • Slack is less and less used, yet they are removing free nonprofit plans

🐦‍⬛ X Hits

  1. Do you find this self-driving action “safe”?

  2. An AI Agent that speaks to car engines

  3. Hyperliquid is the most profitable company per employee, surpassing Tether and OnlyFans

Till next time,
Rares.

Reply

or to participate

Keep Reading

No posts found