Look at me, writing a year-end recap after years of struggling to publish consistently! In 2025, AI stopped being something I wrote about and became central to how I actually write. I only published 9 articles in 2024, but cranked out 47 this year. The difference? I finally built a system: a homemade CMS designed around my actual writing process, Raindrop.io to capture links and quotes effortlessly, and LLMs to catch my non-native English mistakes and typos.
So what have I been writing about this whole time? Surprise, surprise: this year was all about AI. Not the “here’s what AI might do someday” kind. More like “here’s what I’m actually building with it right now.” I’ve been prototyping hands-on, experimenting, breaking things, fixing them (or trying to), and sharing what I learned along the way.
AI-Assisted Coding & Vibe Coding
This was my year of tinkering. I learn by doing, so it makes sense. I wasn’t just reading about AI, I actually used it to ship real (side) projects:
- Built a flight tracker with ChatGPT and Claude
- Created a prompt manager with Bolt and Mistral
- Prototyped a tag manager component with ChatGPT and Cursor
- Tested GPT-5 in Cursor and shipped version history features in 5 minutes (GPT-5 First Impressions)
- Experimented with Codex CLI, AGENTS.md, and PRDs
- Explored Gemini 3 and Google AI Studio for prototyping
The evolution’s been interesting, basically tracking how LLMs and AI tools have advanced. In January, I was just exploring basic AI-assisted coding. By August, I’d moved to sophisticated workflows with GPT-5. And by November, I was experimenting with spec-driven development.
AI Safety & The Lethal Trifecta
As a Product Manager, I know that building reliable and secure products is critical. So I’ve spent considerable time reading (and some writing) about the inherent risks of LLMs and artificial intelligence.
- Simon Willison’s lethal trifecta (private data + untrusted content + external communication)
- MCP security issues (How secure is MCP, really?)
- Slopsquatting attacks (What is Slopsquatting?)
- Bruce Schneier on securing LLMs (We Are Still Unable to Secure LLMs)
- The normalization of deviance in AI (Quoting Johann Rehberger)
LLM Fundamentals & Techniques
I went down a bunch of rabbit holes learning how LLMs actually work and tried to share some of the highlights:
- Chain-of-thought prompting (Street-fighting RAG)
- Fine-tuning vs. RAG (Quoting Devansh)
- Next-token prediction debate (Challenging the “LLMs are just next-token predictors” take)
- System prompts (ChatGPT reveals the system prompt for Tasks)
- Model personalities (Quoting Vincent Schmalbach)
Product Management Meets AI
What kind of PM would I be if I didn’t write about AI’s impact on product management? I wanted to dig into how AI actually changes the work itself, not just as another feature to ship, but as a tool that reshapes how we operate as PMs.
- AI prototyping for product managers (A guide to AI prototyping)
- Domain experts vs. developers (Quoting Drew Breunig)
- AI credit pricing models (Quoting Elena Verna)
- PRDs and spec-driven development (Experimenting with Codex CLI)
- The shift from chatbots to digital coworkers (A First Look at Prototyping With Gemini 3)
Philosophical Takes on AI’s Impact
The impact of AI on humans and the planet is something I want to explore more in 2026. I gave it my best shot in 2025.
- The “introduction of photography” moment for programming (Quoting Orta Therox)
- Idea guys who can just do things (The future belongs to…)
- Coding like a surgeon (Quoting Geoffrey Litt)
- AI replacing jobs vs. augmenting humans (Identify, solve, verify)
- Doomprompting and passive creation (Quoting Anu Atluru)
- Resonant computing principles (Quoting The Resonant Computing Manifesto)
The Numbers
I’m writing mostly to cultivate my digital garden and have a place to reference when I share discoveries with friends or colleagues. But it’s nice to see that some people actually read it.
