Is AI Actually Making You Dumber and Slower?
A shocking METR study finds AI coding assistants make developers 19% slower, not faster.
This AI news report, published on August 2, 2025, covers a shocking METR study revealing that AI coding assistants may actually slow down expert developers. Additionally, Apple signals a major shift towards AI acquisitions to combat its stock slump, the revenue race between OpenAI and Anthropic heats up, and Nvidia faces new security scrutiny from China over its H20 chips.
Slower by Design? METR Study Finds AI Coding Assistants Hurt Productivity
A fascinating new study has just flipped the script on everything we thought we knew about AI and developer productivity. For months we've been hearing that AI coding assistants are revolutionary productivity boosters, making developers faster and more efficient. Well, it turns out the reality might be quite different from the hype. Researchers from METR, an organization that evaluates AI system capabilities, conducted a comprehensive study with expert developers working on real coding tasks. They gave these experienced programmers 246 different issues to fix, everything from bug fixes to new features, and randomly assigned each task to either allow or prohibit AI use.
Here's where it gets interesting. Before starting their work, the developers predicted that AI would make them about 24% faster. That sounds about right, doesn't it? We've all heard similar claims, but when the actual results came in, they told a completely different story. The data showed that AI actually made these expert developers 19% slower, not faster. That's a massive gap between expectation and reality. What makes this even more intriguing is that after seeing the evidence, the developers still insisted they felt like AI had boosted their productivity by about 20%.
So we're dealing with a significant perception versus reality disconnect here. The researchers identified five key reasons why AI might be slowing down professional coders. First, there's what they call the review burden. Developers were spending too much time correcting and fixing AI generated code. Think about it. When AI writes code for you, you still need to understand it, debug it, and make sure it actually works properly. Second, high quality codebases have specific requirements and standards that current AI models simply don't grasp well enough. Third, AI struggles with domain specific knowledge that experienced developers have built up over years. Fourth, expert developers already know what they're doing efficiently, so AI might actually interrupt their natural workflow rather than enhance it. And finally, there's the context switching problem. Constantly jumping between your own thinking and AI assistance can break that crucial coding flow state. Now, it's important to note that this study isn't claiming AI slows down all developers in every situation. What it's capturing is a specific snapshot of early 2025 showing how today's frontier AI models impact experienced open source developers. This could change as the technology improves, or the results might vary significantly for less experienced developers or different types of coding tasks.
Apple's AI Panic: Cook Signals Major Acquisitions to Catch Up
Shifting gears to some major industry developments, Apple is finally feeling the heat from the AI revolution. The iPhone maker's stock has fallen 15% this year, trailing six out of the seven "Magnificent 7" tech companies. CEO Tim Cook has essentially admitted that Apple has sat out the AI wave, but now says the company is ready to open its wallet significantly to catch up with rivals. Cook specifically mentioned being very open to mergers and acquisitions that accelerate our roadmap. There are already reports that Apple executives held internal discussions about potentially acquiring AI startup Perplexity. This is a big deal because Apple has traditionally been very conservative about major acquisitions, preferring to develop technology in house.
The $12B AI Race: OpenAI Dominates, But Anthropic's Growth is Explosive
Meanwhile, in the AI startup world, we're seeing some remarkable revenue growth numbers. OpenAI continues to lead the pack, with a reported $12 billion in annualized revenue and over 700 million weekly active users. But here's what's really interesting. Anthropic, the company behind Claude, has reached 4 billion in annualized revenue and is closing the gap at breakneck speed. Just three years ago, the difference between these companies was about 20 times in OpenAI's favor. Today, that gap has shrunk to just three times. The contrasting strategies tell an interesting story. OpenAI dominates consumer markets with its massive user base, while Anthropic's enterprise first approach is fueling rapid growth in business applications.
Nvidia's China Problem: Beijing Alleges H20 Chip Security Flaws
Another interesting development is happening in China, where regulators have summoned Nvidia representatives over what they're calling serious security vulnerabilities with the company's H20 AI chips. This comes just weeks after CEO Jensen Huang made a high profile visit to Beijing. Chinese officials cited concerns about US lawmakers' comments regarding potential tracking capabilities in exported chips, claims that Nvidia firmly denied. It's another example of how geopolitical tensions are increasingly intersecting with AI technology.
Spotify's Strange Leak: The "Panama Playlists" & A CEO's Bizarre Music Taste
Next up, let's talk about some lighter AI news. There's been some entertaining controversy around something called the "Panama Playlists", an alleged Spotify hack that claims to have leaked the music playlists of major public figures. If true, the strangest playlist definitely belongs to Coinbase CEO Brian Armstrong. It's a seven hour playlist called "Repeat" that features the same song "Long Way Home" by Gareth Emery 60 times in a row. In an amusing twist, Armstrong appears to have confirmed the playlist's existence on social media. Whether this tells us something profound about crypto executives or just that everyone has weird music habits is up for debate.
AI Briefs: Perplexity's New Browser, OpenAI's Privacy Slip & Major Deals
On the practical side, there are some useful new tools worth mentioning. Perplexity's new Comet browser now lets you automate repetitive web tasks like making bookings or comparing products with simple text prompts. And for those interested in no code development, there's been buzz around someone claiming to build a complete AI calorie tracking app in just 20 minutes using various AI tools. Finally, there are a few quick regulatory and business updates. OpenAI appears to have disabled a feature that was accidentally making some ChatGPT conversations publicly searchable, which understandably raised privacy concerns. Google is testing new AI-powered age estimation technology in the US to better filter content across its products. And the New York Times has struck a new licensing deal with Amazon, where Amazon will pay at least $20 million per year to use the Times content for AI training. That's it for today. Make sure to like, subscribe or follow and see you tomorrow for more AI news.
What's Your Take?
The METR study highlights a huge gap between how productive developers feel with AI and their actual performance. Have you experienced this disconnect, or do you find AI coding assistants genuinely accelerate your workflow?
Leave a comment below and join the discussion!
Enjoyed this report? Watch it on YouTube or share it with a colleague.