All will churn for a while. This is a fundamental shift in how we use computers and all the things connected to them. Automate everything is going to occur sooner than later.
One "dark corner" is agent to agent payments. When that takes hold, agents will be paying other agents to do what they are good at, and the ones that accel will make bank and spread.
We are unlikely to survive as-is. We too must evolve.
I believe Agents and the Model Context Protocol (MCP) will remain resilient, while rigid Agentic workflows will likely churn away. Hard-coded, step-by-step state machines—like rigid "if-this-then-that" prompt chains—are just temporary scaffolding. In my opinion, autonomous agents are likely going to be the most resilient piece here, but I also believe they harbor the most unaddressed "dark corners" of the AI ecosystem. A lot of people are rapidly deploying them to multitask like scheduling meetings, running and testing code, etc but there is also a massive lingering issue regarding how much power and execution authority we are giving them, and the unchecked decisions they could make could be problematic depending on the situation.
Not on your list, but getting annoyed with streaming text. Just want to see the message when it is complete instead of getting blasted by large number of tokens rendering per second.