I write about systems that hold up over a decade.
Most tech content optimizes for this week — the latest model release, the framework that shipped yesterday, the framework that will replace it next month. There is a place for that. This is not it.
caphe.dev is for decisions that compound: architectures you will live with for ten years, technology shifts that will redraw industries, choices founders make in 2026 that determine where they stand in 2036.
What I cover
Systems architecture. First-principles thinking over best practices. Why simplicity is the hardest skill in design. Why microservices are usually the wrong answer for teams under fifty engineers. The kind of design decisions that look obvious in retrospect and impossible at the time.
AI as infrastructure, not feature. The interesting question is not whether GPT-5 beats Claude on some benchmark. It is what happens to operations, sales, and middle management when AI becomes a layer underneath every business process — not an app you open. Most enterprise AI projects fail not for technical reasons but because they treat AI as a feature instead of rebuilding the system around it.
AI meeting the physical world. Sensors became cheap a decade ago. Models became capable two years ago. The unlock now is closing the loop — AI that perceives, decides, and acts in physical space. Cold chain monitoring across thousands of stores. Vision in factories that actually ships to production. The shift from screens to ambient computing.
Banking and finance, rebuilt. Banks-as-code is not a vendor pitch. It is what happens when the regulatory and software stacks both become composable at the same time. Southeast Asia has the rare advantage of skipping the legacy systems that anchor incumbents elsewhere. The fintech that wins the next decade does not exist yet.
Engineering in the AI era. Knowing many frameworks used to be valuable. AI can list every framework in five seconds. The skill that matters now is judgment — when best practices do not apply, and how to build the answer from first principles. This is the harder skill. Fewer people are practicing it.
How I write
Long-form, opinionated, specific. I name real systems, real numbers, real tradeoffs. I do not write about news. I do not write about hype cycles. I do not write listicles.
If a post would not matter in three years, I do not publish it.
I write in English and Vietnamese. Same post, different languages — not separate audiences.
Background
Twenty-plus years building systems. The work I am proudest of is open source, in production at scale, and still using the architectural decisions I made years ago.
I have built and led engineering organizations of more than three hundred people, in a market where most playbooks were written for somewhere else. I have shipped AI and IoT systems for enterprise customers, including deployments across thousands of physical locations.
Most of what I write here comes from problems I am working on now, not lessons from a decade ago.
Why caphe.dev
Coffee is where I think most clearly. A black coffee, no rush, no next meeting — that is when the good ideas show up.
This blog should feel the same way. Sit down. Slow down. Think.
AI and how these posts come together
Most posts here come together through dialogue with AI.
Many of the topics I write about — Buddhism, neuroscience, financial architecture, the physics of sensor systems — are not areas I am deeply expert in. I have read enough to ask the right questions, not enough to write alone.
The process usually goes like this. I start from an observation in the field — a technical decision, a meeting with a founder, something that just broke in production. I work with AI to name the phenomenon, pull in adjacent disciplines, stress-test the argument, find the theoretical frame that already exists. I keep what matches my experience and drop what does not. I mix sources that do not usually meet — film with neuroscience, scripture with predictive coding, MQTT with KISS. The final judgment is mine. The final structure is mine. The final sentence is mine.
I do not hide this. I think this is how most quality intellectual work will be done over the next decade — and this blog is a live experiment in doing it that way.
In return, I owe the reader two things: only publish what I stand behind, and write in my own voice — not the average voice of the internet.
Contact
Email is the best way: tuan@caphe.dev. I read everything. I do not reply to everything.
Priority goes to specific questions with real context — preferably from someone building something. Generic outreach gets ignored.