A conceptual hero image for a blog post. An open book lies on a surface with a pen. From the right page, a holographic, futuristic city skyline rises, representing Silicon Valley. A glowing brain icon, symbolizing AI, floats above the city, connected by lines of light. The text "One Year of Building with AI" is on the left page and "Silicon Valley Insights" is on the right.
Reflections

One Year of Letters from Silicon Valley

Allen HutchisonLeave a comment

A year ago, I hit “publish” on the first post for Letters from Silicon Valley, not knowing where it would lead. I just knew I wanted to write about the messy, hands-on process of building with AI. Twelve months later, the blog has become a record of experiments, discoveries, and more than a few wrong turns that taught me something valuable.

When I started, I had a specific itch to scratch. I wanted to see what would happen if I took a large, personally meaningful dataset—in my case, nearly two decades of podcasts—and built an AI system around it. My hunch was that somewhere in those thousands of hours of stories and interviews was a hidden web of ideas I could surface with the right tools. I wasn’t chasing an abstract “AI project”; I was building something for myself, something that could finally answer the question: “Where did I hear that thing about…?”

But there was a deeper reason, too. I had just come off a role managing very large teams and was intentional about what I wanted next. I wanted more time with my hands on a keyboard, building things that were helpful to me and others. I wanted to help developers understand how to use AI in their own work and explain core concepts—like I did in my series on embeddings—in a way that was accessible and grounded in real use. Most importantly, I wanted to reconnect with the simple satisfaction of writing code that solves my own problems.

I found the perfect environment for this on the AI Developer team, working alongside the people behind the Gemini API, AI Studio, Colab, and Kaggle. The team’s maturity freed me from constant hands-on management, giving me the space to build and write. At first, my schedule felt too unpredictable for mission-critical work, so building open-source projects and writing about them was the ideal way to contribute. It also let me lead by example, showing the kind of developer engagement I hoped to inspire.

As the year went on, I grew more comfortable with my schedule and began taking on core projects. The biggest of these is the Gemini CLI, which grew directly from the philosophy of this blog: solve real developer needs through hands-on experimentation.

Along the way, I built and released two other open-source projects:

  • Podcast Rag: Tools for downloading, transcribing, embedding, and searching podcast archives, which grew directly from my initial project.
  • Gemini Scribe: An Obsidian plugin that turns your notes into an interactive workspace with AI-powered recall, summarization, and writing support.

These projects became both proofs-of-concept for ideas I’ve shared here and real tools that others can adopt, modify, and build on.

Looking back, the throughline is clear: building with AI isn’t about replacing what I know—it’s about reframing what’s possible. The best work this year came from combining my own experience with the model’s speed and breadth, letting each push the other in new directions. That’s the spirit I plan to carry into year two.

📅 Year One Recap

September 2024: The Podcast Project

My first posts documented the journey of turning two decades of listening into a searchable, AI-powered personal knowledge base.

October 2024: The Embeddings Series

A deep dive into one of the most important—and misunderstood—concepts in modern AI, showing how I used them in the podcast project.

November 2024

December 2024

March 2025

April 2025

May 2025

June 2025

July 2025

August 2025

What’s Next

Year two will bring more of the same, just with more intention. I started year one with an ambitious goal of publishing weekly, but quickly learned that pace wasn’t sustainable. This year, I’m aiming for at least two thoughtful posts a month—a balance that allows for a regular cadence and the breathing room to go deep.

The explainer series on embeddings was some of the most-read content I published, and I plan to do more of that. There are plenty of concepts in modern AI that could use the same treatment, and it’s a great way for me to learn alongside my readers.

At the same time, not every post needs to be a 10,000-word essay. I plan to mix in more short pieces—quick takes, reactions, and smaller ideas worth sharing in the moment. I enjoy the variety on Simon Willison’s Weblog, where short links and deep dives sit comfortably side-by-side, and I’d like to bring more of that spirit here.

When I launched this blog, I imagined it covering more than just AI. My “About Me” page mentions guitar building and woodworking, but I’ve hesitated to bring those topics in, worried they might feel “off-topic” for those who subscribed for AI content. This year, I’m going to embrace it. Those interests are part of the picture, and they may connect to the technical work in ways I don’t yet expect.

Finally, I’ve been inspired by the personal, thoughtful writing of former colleagues like Brian Brown (Changing Coordinates) and Chris DiBona (Substack). I hope to try a few posts in that spirit as well, sharing ideas that don’t fit into a narrow category but feel important to explore.

Leave a Reply