A year ago, I shared some reflections on how I was using AI and suggested that it’s helpful to think of these tools as competent interns working remotely: earnest and sophisticated, but still in need of direction and supervision.
In 2025, those interns grew up. What surprised me wasn’t the pace of technical progress but how quickly AI stopped feeling novel and became ordinary. I didn’t realize how deep the integration went until OpenAI’s year-in-review revealed I had logged over 3,500 chats, placing me in the top 0.1 percent of users, apparently edging out Sam Altman.
My “tech stack” includes ChatGPT, Gemini, Claude, Manus, Perplexity, Google’s AI Studio, NotebookLM, ElevenLabs, MidJourney, Elicit, and Claude Code. With this context, I felt it might be useful to share how I used these AI tools to tap into “expertise” this last year.
AI as a Research Analyst
I regularly use AI to analyze academic papers, legislation, and reports, not just for summaries but also to explain statistical methods I haven’t touched in years or to pressure-test a study’s methodology. Gemini has also helped translate Chinese AI policy documents.
Google’s NotebookLM has become especially valuable. I maintain topic-specific notebooks (e.g. AI Action Plan, a Congressional hearing) that let me work with only curated sources. You can also take the resources and turn it into a podcast, presentation, or infographic.
Deep Research capabilities have matured across all models. The key insight: these systems perform best when treated like consultants. I now frame prompts as a formal scope of work, defining the task, context, specific questions, and expected deliverables. The results, often 30 pages, have been invaluable for complex medical information, market research, and policy analysis.
AI as an Advocate
When my car’s AC failed, a technician diagnosed it as a leaking evaporator core—an expensive repair. Before approving anything, I uploaded the technician’s report, diagnostic readout, and leak test video to Gemini. It confirmed the diagnosis, validated the estimate, and walked me through the insurance claim process. This is the kind of task that used to require a mechanically savvy friend; AI gave me that expertise on demand.
I also used ChatGPT to successfully dispute a parking ticket and a fraudulent phone account opened in my name. These aren’t glamorous examples, but they represent real value: AI as a personal advocate in asymmetric situations where institutions have more information than you do.
AI as a Co-Worker
Email and calendar integrations are still early, but hint at a different future. Features like ChatGPT’s Daily Pulse and Gemini’s Agent tools provide lightweight briefings that synthesize my calendar, flag new studies, and summarize relevant news.
When these models connect to your email and files, they stop being standalone chatbots and start operating like a co-worker. Previously, using AI meant pulling your work out of its natural context and feeding it into a chatbot. Now the intelligence comes to where your work already lives, quietly flowing through it instead of demanding that the work move elsewhere.
AI as a Creative Director
Image generation improved dramatically this year. I continue to regularly use Midjourney, Gemini Nano Banana, and ChatGPT Images to create stock art and editorial-style illustrations for blog posts and presentations.
In one instance, I wanted to replicate the look of a photographer I admire. I shared reference images with Claude and asked how to achieve that aesthetic in Lightroom. Within seconds, it broke down the look into component parts—tone‑curve adjustments, HSL shifts, and calibration settings. Even better, it generated a downloadable Lightroom preset, turning a complex aesthetic analysis into a one‑click edit.
AI as Junior Engineer
This was the year AI became a real coding partner. If you have ever thought, “I wish there were a simple tool that did X,” the shift is that you can now just build it yourself. Using Claude, I created an app that compares conference attendees across several years to identify repeat organizations and individuals. With Gemini, I built a tool that filters and summarizes newly published AI research from arXiv.
AI as an Event Manager
Starting from a prior reception invoice, I gave ChatGPT my budget, guest count, menu options, and restaurant contract. It approached the problem the way a seasoned event planner would: extracting per-person benchmarks, modeling menu tradeoffs, estimating service fees, and iterating until food, bar, and logistics fit within the budget.
But the moment that changed how I think about AI entirely did not happen in an office or a spreadsheet. It happened when I began using these tools to help my mom navigate her recurrence of cancer. That experience deserves its own reflection, and it reshaped my understanding of what it means for AI to truly show up in moments that matter. More on that in the next post.