Keerat Rashid

AI agents no longer just answer questions — they book meetings, approve refunds, call APIs, update records, and chain together dozens of small decisions into a single outcome. Most of the time, this works quietly and well. But when something goes wrong, a troubling question surfaces: why did the agent do that? For a large share of deployed systems today, there's no good answer. The reasoning, the data consulted, the tools invoked, and the intermediate steps simply weren't recorded.

I was on twitter the other day, and apparently there's a category of internet content that shouldn't work but does. It's AI-generated, it's obviously AI-generated, and somehow that's the whole point. A cursed content, brain rot, digital slop, but often a piece of it breaks through the noise and becomes genuinely, inexplicably great. "AI slop" became internet shorthand for the flood of low-effort, mass-produced AI content clogging feeds: the fake product photos, the uncanny "motivational" images.

In February 2025, AI researcher Andrej Karpathy described a new way of writing software: describe what you want in plain language, let an AI model generate the code, and largely stop reading it line by line. He called it "vibe coding." A little over a year later, it's not a novelty — it's close to the default way software gets built.