Stack Overflow: 84% Devs Use AI, but Only 29% Trust It

Inside: Popular npm Hacked and Alibaba vs Google & OpenAI

Author

Chris Lojniewski
July 30, 2025

Hello!

OpenAI is thinking about school already. They’ve introduced a new Study Mode, just as Cursor updates to 1.3. 

However, we’re getting more signals that devs are getting frustrated with AI. The latest Stack Overflow Developer survey shows that trust in AI tools fell to a low 29%, even though 84% if devs use them.

MIT released research showing how ChatGPT is affecting our brains, and Anthropic plans to introduce new rate limits in August, despite user protests.

And for something completely different, you can learn how to track airplanes with Grafana!

Grab your coffee and enjoy this issue of Frictionless!

Chris Lojniewski | Pagepro LinkedIn

Wanna chat about the newsletter or have questions for me? Let’s connect!

www.linkedin.com/in/chris-lojniewski

In the Queue

Reduce Friction

A horizontal bar chart titled “More developers actively distrust the accuracy of AI tools than trust it.” Trust levels among developers are shown as: Highly trust: 3.1% Somewhat trust: 29.6% Somewhat distrust: 26.1% Highly distrust: 19.6% The visualization highlights the skepticism developers have toward AI tool accuracy.

Source: Stack Overflow

Developers Remain Willing but Reluctant to Use AI: The 2025 Developer Survey Results

80% of developers are experimenting with AI tools, but trust in it fell from 40% to 29% this year. Stack Overflow’s reveals a reason behind this growing divide: debugging. That’s only one insight from the survey, so if you’re curious how many devs use agents or are happy at work, read the results.

Software Architecture Is Finally Fixing Its Biggest Problem: Developer Experience

Treating architecture like a product, and not only a backend concern, pays off big time. Modern teams at top companies are taking DX into account when designing systems and see serious productivity gains. 

How to Work with AI: Using AI as a Thought Partner

Are you using AI effectively? Having it only execute tasks is a waste of potential. After all, it can be the perfect collaborator to you and your dev team, who will sharpen your thinking and open up new perspectives… provided you’re not overrelying on it.

🎥 How Vercel Saves Dev Teams $10M+ in Productivity Gains

Don’t miss this interview with Guillermo Rauch, the CEO of Vercel. Inside, Guillermo goes over how his mindset helped Next.js take off and turn Vercel into the empire that it is today. You’ll leave with sharp insight into product thinking and how to support developer speed.

Deepen Your Expertise

A tweet by developer JouQin tagging ESLint and Prettier maintainers. He discloses that a phishing email led to a leaked npm token, which resulted in malicious versions of popular packages being published. He states that the token has been revoked, malicious versions deprecated, and new secure versions released.

Source: X

How npm Package with 30 Million Downloads Spread Malware

One of the most trusted packages in the JavaScript ecosystem,  eslint-config-prettier, was hijacked and used to steal developer tokens. A rogue actor slipped through weak 2FA and npm’s verification loopholes, turning a routine dev tool into a credential-harvesting trap.

How Cursor Serves Billions of AI Code Completions Every Day

Have you ever thought what it takes to deliver 1+ million queries per second on AI code suggestions? Look under the hood of Cursor’s infra and learn how it handles fast token streaming, encrypted code isolation, and cross-cloud failover. And speaking of Cursor…

Cursor 1.3 Is Here

Cursor's new update now allows agents to operate directly in shared terminals and edit up to 25% faster! Have you tried it out yet?

A flowchart titled “The ‘Hill Climbing’ Approach to Improving Models” illustrates an iterative process for machine learning model refinement. The chart is divided into three main steps: Baseline Evaluation (blue box): Action: Evaluate current model Icon: Upward-trending graph Analyze Failures (beige box): Action: Identify weaknesses Icon: Magnifying glass over an “X” Re-evaluate (green box): Actions: Make a targeted change (illustrated by a light bulb) Assess new version (illustrated by a hill with an arrow pointing upward) An arrow connects each step in a linear flow, ending with a looping arrow labeled “REPEAT”, showing the process is cyclical. The goal is to continually "step up" the model’s performance through small, focused improvements.

Source: Addy Osmani Blog

An Engineer's Guide to AI Code Model Evals

Addy Osmani, engineering leader at Google, shares how to evaluate coding models with real-world reliability. His guide covers golden tests, eval automation, and avoiding overfitting traps, so give it a read if you’re trying to improve your own AI workflows.

Contentful Migration in 2025: What are the Alternatives?

React components shouldn’t take over 500 lines, but that was our client’s reality with Contentful. If you, too, are struggling with its complexity or rising costs, I’ve prepared a list of Contentful alternatives, which will make your team (and wallet) happy.

AI Corner

A scientific-style visual titled “Your Brain on ChatGPT” displays multiple 3D brain diagrams overlaid with EEG Alpha Band connectivity data (using dDTF values). Each diagram highlights different neural network activity patterns during tasks involving: LLM (Large Language Model) use – marked with red arrows Search engine use – marked with green arrows Baseline brain activity (non-AI task) – marked with blue arrows The connectivity lines between brain regions are color-coded: From blue (lower dDTF) to red (higher dDTF) on the left-hand scale And green to orange based on p-value significance on the right-hand scale The image implies distinct differences in cognitive activation patterns depending on the tool being used, with LLMs, search engines, and baseline brain activity showing varied connectivity signatures. The upper-left corner includes a grayscale drawing of a person interacting with a computer, reinforcing the experimental context.

Source: BrainOnLLM

Your Brain on ChatGPT - MIT Study

MIT researchers used brain scans to see how ChatGPT affects our thinking. Participants not only had trouble recalling what they wrote with AI’s help, but also had their brain connectivity drop by 47%, from 79 to 42 points.

Alibaba’s Qwen Tops OpenAI & Gemini 

Alibaba's Qwen3 model is beating GPT-4 and Gemini 2.5 on core reasoning tests, all while being open source. Google and OpenAI might want to watch their back for now, but will Qwen3 hold up against the upcoming GPT-5?

A screenshot of a tweet by Eric Hartford replying to Anthropic. Eric announces that he’s cancelled his Claude subscription after being a long-time user, reacting to Anthropic's tweet about new weekly rate limits for Claude Pro and Max subscribers. Anthropic’s post estimates that fewer than 5% of users will be affected by the changes. Eric’s message expresses disappointment, implying frustration with the decision.

Source: X

Anthropic Unveils New Rate Limits to Curb Claude Code Power Users

Anthropic is preparing to introduce new weekly rate limits for Claude Pro and Max. The claims that this is caused by “fans running Claude continuously in the background, 24/7” met with sharp criticism and accusations.

OpenAI Announces New ‘Study Mode’ Product for Students

ChatGPT might join the list of student essentials this year. Study Mode is OpenAI’s latest attempt at structured usage. It's a prompt-guided assistant built for learning, but it might come in handy for more than just school chores.

Just Cool

A photo of a large flatscreen monitor mounted on a wall showing a live air traffic monitoring dashboard (likely using Grafana). It features a map of the UK and Ireland with green dots representing aircraft. Sidebar panels show aircraft data, including current count, airline types, and status (e.g., “No Emergency Aircraft”). Brightly colored summary boxes display flight counts, such as "EasyJet Flights: 7" and "Ryanair Flights: 12."

Source: Grafana Blog

Tracking Planes With Grafana in Real Time: How to Visualize the Aircraft Overhead with Your Own Dashboard

Yes, Grafana can track planes overhead in real time, and no, it’s not just a gimmick. If you love airplanes, learn how to set up a tracking system using open aviation data and live dashboarding.

Let’s Stay in Touch! 📨 

Do you have any comments about this newsletter issue or questions you want to ask? Drop me a message or book a meeting.

What do you think of today's email?

Your feedback helps me improve Frictionless.

Login or Subscribe to participate in polls.