2025 Year in Review: The Quiet Power of Steady Progress
My 2025 AI journey boils down to four numbers:
These 6 speaking engagements included 1 DevOpsDay, 4 iThome Hello Worlds, and 1 Google DevFest Taipei.
For me, 2025 is a year worth reviewing — not because it was incredibly glamorous, but because it felt incredibly grounded. It was like finishing a little bit every day, and only when I looked back did I realize: wow, I’ve walked this far.
GenAI in Banking: Replacing Plumbing in an Operating Building
I’ve always felt that deploying GenAI in a bank is somewhat like replacing the plumbing in a building that’s still fully operational.
You can’t cut the water or power, you can’t disturb the tenants, and you absolutely cannot let anyone feel like “this modification is dangerous.” Because of that, the truly difficult part is rarely the model itself. The real difficulties are engineering, workflows, governance, boundary setting, accountability, and a massive pile of unromantic but critically important details.
In 2025, I became even more certain of what I love doing: taking AI — which often seems “mystical” — and turning it into a capability that can be verified, trusted, and operated long-term.
Patents: A Process of Forcing Honesty
The patent process is similar.
Getting 6 utility patents granted with 5 invention patents pending isn’t just about collecting certificates for me. It functions more like a mechanism to force honesty upon myself:
- What problem did I actually solve?
- Where are the boundaries of my proposed method?
- Can I articulate the value in a language that others can read and verify?
Every time I finish writing one, it feels like I’ve just gone through a brutal code review — except the reviewer isn’t a colleague, it’s the world.
Then there’s the conference paper submission, which held a lot of symbolism for me. It wasn’t just about throwing an idea out there; it was about putting myself in an arena where I would be challenged and corrected.
I know very well: if I don’t write it down, it will forever remain a “I feel like…” thought. But once it’s written, it becomes something that can be debated, improved, or even overturned.
Strangely, I actually enjoy that uncertainty — because it means I’m still learning, still growing.
Speaking & Sharing: An Arena for Self-Verification
As for speaking, I really did a lot in 2025:
- DevOpsDay (1 session)
- iThome Hello World (4 sessions)
- Google DevFest Taipei (1 session)
I get nervous before every time I take the stage. But the nerves aren’t about speaking poorly; they’re about sounding too much like a slogan, too much like “AI is amazing, everyone get on board.” I prefer to talk about reality: the truly difficult part of adopting GenAI in an enterprise is almost always how you make it controllable, traceable, maintainable, and verifiable.
The more you speak, the more you realize that being able to explain something clearly is a form of self-verification. The parts you can’t explain clearly are the parts you haven’t fully thought through yourself.
2026: Run, don’t walk!
What I learned in 2025 wasn’t how to be faster, but how to be steadier.
It is too easy to be dragged along by speed in this era. But I increasingly believe: what truly holds long-term value is quiet, solid execution — making systems stable, clarifying workflows, designing for risk, and putting people before technology.
These things aren’t flashy to build, but once they’re done, they foster a very reliable kind of confidence.
If you are doing the same thing — turning technology into something that helps people in an environment with high constraints and heavy responsibilities — I want to tell you that you’re already doing something incredibly difficult. You don’t need to move the whole mountain at once; you just need to push a little bit more every day. Those seemingly small accumulations will ultimately become your most reliable weapons.
Thank you, 2025. And thank you to the great partners who worked alongside me and were willing to exchange ideas along the way.
In 2026, I will keep running.
Run, don’t walk! Either you’re running for food, or you’re running from being food.
#YearInReview #GenAI #Patents #Research #DevOpsDay #HelloWorld #DevFestTaipei #AIGovernance #Productization
💬 Read more: 2025 Year in Review (English)
Are You Ready to Become an AI PM? A 5-Dimension Self-Assessment
This isn't a quiz about RAG or prompts — it asks: in real AI product scenarios, is your judgment ready? A 20-question skills check across 5 core dimensions to map your AI PM readiness.
When All the Roles Merge: Reflections on PM in the AI Age
Anthropic's Cat Wu describes a new PM rhythm in the AI era: roles merging, prototypes over docs, iteration in days not months. Reading it brought back memories of my own undefined role in an enterprise AI team—and Peter Deng's Avengers-style team philosophy.
Are You Ready to Become an AI PM? A 5-Dimension Self-Assessment
This isn't a quiz about RAG or prompts — it asks: in real AI product scenarios, is your judgment ready? A 20-question skills check across 5 core dimensions to map your AI PM readiness.
When All the Roles Merge: Reflections on PM in the AI Age
Anthropic's Cat Wu describes a new PM rhythm in the AI era: roles merging, prototypes over docs, iteration in days not months. Reading it brought back memories of my own undefined role in an enterprise AI team—and Peter Deng's Avengers-style team philosophy.