I don’t know about you, but I am so tired of the “AI revolution.”
Remember 2023 and 2024? You couldn’t read a single article, watch a single video, or even buy a smart toaster without being told how “Generative AI” was about to change all of human existence. Every startup was an “AI company,” and every “guru” on social media was five minutes away from achieving Artificial General Intelligence in their garage. It was a gold rush. It was a bubble. And it was, frankly, exhausting.
Well, we’re at the end of 2025, and I’m happy to report that the bubble has burst. This was the “Great AI Hangover” year—the year we collectively woke up, looked at the mess, and finally started cleaning it up.
And as a gear-head, a tech editor, and someone who actually likes making things, I can tell you this: it’s the best thing that could have happened. The hype is dead, and the utility is finally here.
The “Do-Everything” Dream Is Dead
The first casualty of 2025 was the “do-everything” model. The race to build the biggest, most god-like Large Language Model (LLM) hit a wall of reality. Sure, the new models are better, but they weren’t the leap to sentience everyone breathlessly predicted.
Instead, the market finally figured out what some of us knew all along: a “jack of all trades” model is a master of none.
The real revolution this year wasn’t in the cloud; it was in what the industry is calling “Vertical AI.” Instead of one giant AI trying to write a poem, diagnose cancer, and trade stocks, we now have specialized models. We have AI for healthcare that’s trained only on medical journals. We have AI for construction that only understands blueprints and material stress.
This is the “boring” AI that’s actually changing things. It’s not a party trick; it’s a tool. It’s the difference between a flashy concept car and the high-torque wrench that lets you build a real one. The endless flood of “GPT-wrappers” (thin apps that were just a skin on an API) are gone, and in their place, we’re getting software that has AI baked in to do one specific job incredibly well.
The Real Power Is in Your Hand
The second, and for me, more exciting, shift is that AI is finally moving out of the data center and into our hardware. For years, “AI” meant typing into a box that was connected to a billion-dollar server farm on the other side of the country. It was slow, and it was a privacy nightmare.
2025 was the year of on-device processing.
We’ve been covering the hardware at Modded for a long time, and this is the payoff. Companies are finally in a silicon arms race over their NPUs (Neural Processing Units). When Apple announced its M5 chips earlier this year, they didn’t just talk about faster CPUs; they bragged about Neural Accelerators in the GPU and a faster Neural Engine.
Why does that matter? Because it means the AI runs on your laptop or phone. It’s fast. It’s private (your data isn’t being sent to a server for “training”). And it’s personal. When your device learns your habits, it’s not to sell you ads; it’s to make the tool work better for you. This is the AI that gives you real-time photo editing without a “processing” spinner, or a smart assistant that actually knows your context without being a creep.
This move to “Small Language Models” (SLMs) that can run locally is the single biggest leap toward making AI a practical part of our gear, not just a service we subscribe to.
AI Finally Gets a Body
The last piece of the puzzle is the one I care about most: the AI is no longer just a chatbot. It’s getting a physical body.
We’re seeing it in the cars we’re testing. The ADAS (Advanced Driver-Assistance Systems) in 2025 and 2026 models are on another level. They’re not just “lane-keeping” anymore. They’re using these new, efficient AI models to process camera and lidar data in real-time, allowing them to understand complex, messy urban environments. It’s still not “self-driving,” thank God—I like to drive—but it’s a co-pilot that’s genuinely smart and not just a nervous backseat driver.
We’re also seeing the rise of “AI Agents”—software with the power to do things. Not just suggest, but act. In the design world, engineers are telling an AI, “Run 1,000 stress tests” on a part, and the agent does it. In coding, agents are making their own code changes and submitting them for human review.
This is the future I’m here for. The hype was all about AI replacing us. The reality is that AI is becoming a powerful, specialized, and physical tool that helps us do our jobs better.
So yeah, 2025 was the “Trough of Disillusionment,” as the big analyst firms call it. The party’s over. But for those of us who actually build, drive, and use tech, the real work has just begun. And it’s going to be a hell of a lot more interesting than a chatbot.



