AI tool decreases company cognitive function.
Also, CPU metrics are wrong and Garmin shipped a watch.

The Mandatory AI "Upskilling" Course Also Destroys Memory Retention

A new study from MIT has formally confirmed what every seasoned Systems Administrator already knew about their junior hires: the pervasive use of LLMs is actively making people stupider. The research, which employed EEG scans on participants, found that heavy use of the tools weakens neural connectivity, leading to a demonstrable cognitive decline and impaired memory recall. This is not a bug; it is a feature designed for maximizing shareholder value by achieving peak "cognitive offloading," allowing the human to be nothing more than a slow, fleshy API endpoint.

The most encouraging detail is that students using the AI had a reduced sense of ownership over their work. This means the next wave of layoffs will be even easier, as the employees being dismissed will be unsure if they wrote the code or if the prompt wrote the code. Management can simply cite the reduced sense of ownership as a perfectly reasonable performance metric. The study also noted that brain activity remained below baseline even when participants stopped using the AI, suggesting a permanent adaptation toward passivity. The new corporate brain is now optimized for clicking "Approve" and nothing else.

Where Are the Bad Apps: The Great AI Productivity Audit

Everyone has been loudly assuring the quarterly report readers that AI has made engineering 10x more productive. Unfortunately, this claim presents the new AI Fermi Paradox: if every engineer is 10x faster, where is the flood of garbage software, the low-effort indie revolution, the tidal wave of new apps we expect from a massive technological leap? The answer, according to one veteran, is that the productivity gain is entirely internal, allowing developers to finish the personal projects they never had time for before.

Instead of shipping more products to the market, the AI has simply allowed engineers to ship more things to themselves. This is a crucial distinction. The company bought a new Ferrari for the entire engineering department, and now they are just driving it around their own private backyard. Furthermore, one developer pointed out that while they felt 25% faster, an older study suggested AI was actually making them slower, proving that developers are unreliable narrators of their own performance. Never trust a dev who self-reports an efficiency boost.

Yes, Your Dashboard Is Lying to You, Says Someone Who Reads the Manual

We now know that %CPU utilization is a lie, which should be immediately printed and stapled to the forehead of every person who has ever yelled at Operations about a "spiking metric." The venerable metric is an ancient compromise, an easy number for managers to point at, but it rarely reflects what the underlying hardware is actually doing. The article explains that the percentage is often a measurement of the CPU's "idle time"—a number that is then inverted, rounded, and generally beaten into submission until it fits a nice graph.

A good metric, it turns out, is hard. A number you can confidently yell about in a weekly stand-up is easy. This news will undoubtedly be ignored by every executive whose primary job function is to scan the monitoring dashboard for anything red and send an urgent, all-caps email about it. They will keep yelling at the inverse of the idle time until the actual problem fixes itself, which is often how most IT problems are solved anyway.

Briefs

• Emulation: The 16-year odyssey to emulate the Pioneer LaserActive has ended. Congratulations to the person who now owns the most difficult-to-launch 1993 hardware in existence.
• Smartwatch Arms Race: Garmin beat Apple to market with a satellite-connected smartwatch. Apple will now spend $100 million on a marketing campaign explaining why they chose to release their version last, calling it "thoughtful prioritization."
• Microsoft History: Microsoft released BASIC for the 6502 Microprocessor, version 1.1. Finally, the next generation of coders can learn to write games that use exactly 4KB of memory, provided they can find a 6502 machine to run it on.