Sumiya

On a Pro subscription for Claude, and last night I was able to send just a single message. “Continue this conversation”. And then it went off and kept working until it exhausted my session limit. That was around 10 PM. It is about 5 AM now and I sent “Continue now.”. Let’s see what happens next.

Checking back after work. The next prompt was "Continue now.". It used 40% of the allowed limit. I am beginning to think I am doing something wrong.

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

None of these are permanent advantages. They are simply skills that will hold their value longer than syntax knowledge or pattern matching. 26/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

This knowledge is almost never written down anywhere useful. It lives in relationships, in institutional history, and in reading a room. 24/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

Context matters! The last important thing I want to say is that a lot of engineering judgment in an enterprise is actually about understanding the environment a technical decision has to survive in. 22/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

AI makes the path of least resistance so much smoother that it becomes dangerously easy to stay in your comfort zone. I practice working in unfamiliar territory deliberately: not to become a generalist, but to practice the specific skill of reasoning without reliable priors. 20/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

Be a beginner! The thing I find most uncomfortable, and arguably most vital, is placing myself in contexts where my experience is a weak guide. 18/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

But I want to be careful about overclaiming here. AI has been trained on all of that literature, too. The edge, if there is one, is in the "situated recognition" of when a particular analogy actually applies to a specific problem I am standing inside. 16/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

Here are some things I have tinkered with recently: - Built iOS App without knowing Swift: sumiya.page/2026/Februar... - Mapped the entire Java programming language to command line: sumiya.page/2026/January... - In browser option trade insights: sumiya.page/2025/Novembe... 14/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

Think across domains - AI won't do that yet unless prompted! I treat my curiosity as a research strategy. Steve Jobs famously said you cannot connect the dots looking forward; you can only connect them looking backward. 12/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

The human advantage here is accountability. AI can be a partner in brainstorming potential failure points, but it doesn’t have skin in the game. It can suggest five ways a system might fail, but it doesn’t have to live with the fallout of that decision. 10/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

A machine is not accountable - use that to your advantage. AI is brilliant at "given this problem, here is a solution." It is less reliable at "is this actually the problem we should be solving?" 8/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

Post-mortems exist, but they describe what failed, not what it felt like to make the decision from inside it, with incomplete information and organizational pressure in every direction. Most of these insights never reach the public, and even those that do are often sanitized. 6/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

Learn from failures! AI models learn from documented outcomes. Code that shipped. Architectures that scaled. Decisions that turned out right. This makes them powerful at recognizing patterns from known solutions. 4/

Assume that the optimists are right about AI. Then look around in your environment and see where you add the most value, then introspect how would AI disrupt that. I did some reflection and wrote a post. Thread below. Or read on my blog: sumiya.page/2026/March/s... 1/27 #ai

The question I kept asking myself was which AI tools to use and how to use them well. That matters, but it is probably the wrong place to spend most of my attention. 2/

It’s fun building stuff for myself. It still takes effort but I can push in unknown territories with AI. Built an iOS app for myself for the first time ever. Over the past few evenings and weekends, I built OctaneLog to help me record my story - an iOS app that turns my daily commutes … 1/

… and road trips into vivid narratives, automatically capturing and interpreting the world rushing past my windshield while I keep my eyes where they belong: on the road. And I don’t know Swift. I’ve never built an iOS app before. Not available on the App Store but you can build and install … 2/

Today in things kids say. The other day I was trying to work with my daughter on something and I ended up saying: “This is how I used to do this when I was a kid.” 1/3

And with all her innocence and the seven years of wisdom she has, she waited a couple of seconds, took a deep breath, gave me her c’mon now look, looked deep into my eyes, and calmly said to me - “but Daddy, I am not you. I can do it my way”. 2/3