Weapons of Mass Distraction Are Amongst Us
And it's not the robots.
We’re less than one year into the Agentic AI reality and already the gloss is wearing off the dream, and the reality is looking grim. The problem is that Agentic AI is actually democratizing access to functional AI, AI that does a lot of different things in the digital world. Forget the horror stories about people losing money when they give agents access to their wallets. That’s not the real problem.
The problem is that all of a sudden coding is easy. So easy that anyone can code. So easy that software engineers are expected to 10x or 100x their output. So easy that everyone expects 10x or 100x more software. For everything.
What worries me is that no one is going to have the time to create good software anymore because they are too busy fixing all the AI generated slop. Agentic AI is the worst weapon of mass distraction.
Garbage in, garbage out.
There is almost no cost in creating code but there’s still a huge human cost in checking, troubleshooting or fixing code. This keeps flowing downstream. First it’s burnt out engineers and an abundance of bad code. Now, companies are starting to count the cost of letting Agentic Ai loose anywhere in the stack.
Sadly, it’s really too late. Agentic AI was trained on all the code on the internet. That’s not good. That’s a lot of very bad code. Good code is short and sweet so proportionately there’s a lot less good code online than there is bad code. And there’s been no incentive to get rid of all of those beginner Github repositories.
When code moved at the pace of software engineers, you might have to search past the first few answers on Stack Overflow but you didn’t have to search thousands of answers. Github is starting to go down under the weight of agentic pull requests.
Sure it’s an intellectually interesting solution. Certainly there will be a few crawling out of the proteanic slime. But we need to see evidence of fitness, of survival and utility in the real world over a period of time.
In the meantime, all the code on the internet is now 100x worse than it was just one year ago. This is one instance when more data will not solve the problem.
For me, it’s one more reason why robotics is the best part of the AI revolution.
The real risk is not “AI becoming intelligent”, but low-friction automation creating massive downstream quality, security, and maintenance debt. Robotics is different precisely because the physical world still imposes discipline: things must work, safely and repeatedly, under real constraints. That may make robotics one of the healthier frontiers of AI — slower, but more accountable.