Sensory Overload, 2026

It’s hot and humid this morning in Hong Kong. The rainy season is here, and the air is heavy and wet. A storm is coming as I embark on my commute. I’m trying to stay cool and collected. This Monday morning, I’m working pretty hard to calm my worried brain.

I’m sitting in my usual spot on the ferry; there are only a few rows of seats where you can avoid looking at the giant flashing screens. The displays repeat the same three-minute videos about cybercrime and why you should visit Discovery Bay. If I have them in my peripheral view, I cannot avoid looking at the flashing images; the reptilian brain senses danger.

Now in the second row, I am cooling down and thinking through how I might proceed with a challenging week. Suddenly, I hear the unmistakable sound of somebody dropping a phone. Why is it so easy to recognize as a sound? It’s two sounds in one: first, the collision point hitting the floor with a sharp, short sound, then the lower-pitched, slower heft of the flat surface, hitting the floor. Crash and thump, in short succession. These days, it’s not a big deal; most phones survive the fall, but I think by now, we are all conditioned to have the “did the screen break” question hit our brains immediately.

About 99% of the commuters are on their phone. I start reading a novel in the Kindle app; the book is called “Transcription”. In the first chapter, our protagonist drops and shatters his phone, setting the drama in motion, as he intended to use it for important work. The protagonist’s problem wasn’t the broken screen. It was the dependency. I uneasily put my phone away and fussily open my paper notebook. I’m scribbling down a few thoughts and to-do items. Today, I need to research the Singapore AI regulatory framework, review where my team stands across different client pursuits, and prepare for this week’s big AWS event here in Hong Kong. Clearly, some of this work could be done by my highly proficient but slightly ominous AI assistants. Unfortunately, I burned through 70% of my tokens in the first four days of the month, so I’m parking a lot of things that could be delayed till the end of the month. I can do them just before the token balance is reset. I need to keep at least some tokens for emergency tasks, and I can find a manual way to handle low-priority stuff or drop them altogether.

I have some other personal subscriptions outside the work token budget, but I have a large corpus of knowledge embedded in my Cursor environment, and it would be a pain in the ass to figure out how to efficiently migrate it to other environments to do the work. I’m also in the process of generating reusable code to automate steps that don’t require dynamic intelligence, and I hope to reduce token usage by executing certain tasks as simple scripts.

The overnight news that Cursor is shutting down Frontier LLM access to Mythos/Fable to comply with the US government mandate is a good wake-up call. I need to keep my stuff portable and nimble, and so should we for our client work. It’s convenient to think about LLM access as just another piece of infrastructure, like your cloud service provider or your SaaS CRM vendor. But it’s not the same; this technology and its vendors are rather unpredictable and fickle when it comes to access, quality, or pricing.

Arriving in Central, I step off the ferry; the heavy, hot, humid air assaults me right away, and I scurry for the comfort of the air-conditioned office. It’s a quiet, cool, and calm environment, and I am ready to face the day’s challenges. I open the paper notebook and review the priority tasks. The paper texture feels pleasant to the touch.