My AI Toolkit (Dec 2025)

A few years ago, the "AI toolkit" conversation was mostly about which single chatbot you were subscribed to. Today, it’s becoming more and more like building a specialized team. I’ve found that trying to force one tool to do everything—write code, design architecture, and deep-dive research—is a recipe for frustration.

I’ve been refining my own stack lately, mostly because I’m trying to reduce the friction between "having an idea" and "shipping the thing." (As I’ve written about before, my brain loves to find reasons to fret and stall; good tools remove those excuses.) Here is the current snapshot of my AI toolkit and, more importantly, how I’m using each piece to actually get work done.

The Engineer: Claude Code + Opus 4.5 For pure full-stack and AI engineering work, this is my daily driver. I’ve shifted almost entirely to Claude Code backed by Opus 4.5. There is a specific texture to the way Opus 4.5 handles complex engineering context that I haven’t found elsewhere. It doesn’t just autocomplete; it understands the intent of the architecture. When I’m in the weeds of a React component or debugging a Python backend, I don't want a "chat." I want a co-engineer who has read the documentation I was too lazy to read. Opus 4.5 seems to have that level of context retention that makes the "coding partner" metaphor finally feel real.

The Strategist: Gemini Pro While Claude handles the code, I’ve found myself using Gemini Pro more and more for the "30,000-foot view" tasks. Strategy documents, high-level architecture planning, and especially generating graphics—this is where Gemini shines. There is a fluidity to its reasoning capabilities that works incredibly well for structured thinking. If I need to turn a messy brain-dump into a coherent strategy doc or visualize a data flow, Gemini Pro is the first tab I open. It feels less like a code-generator and more like a clever PM who helps me organize my chaotic thoughts into something shippable.

The Researcher: ChatGPT 5.2 (Pro Mode) Old habits die hard, but for good reason. I still keep ChatGPT (running 5.2 in Pro mode) in the rotation, specifically for long-running research tasks. When I need to go down a rabbit hole—whether it’s understanding a new library ecosystem or researching a niche AI implementation—ChatGPT 5.2’s ability to maintain a thread over a long research session is unmatched. It’s my "deep dive" tool. It doesn't hallucinate the small details as often, and it seems to grasp nuance in research queries better than the others.

The Shift: Coding on the Go The most interesting shift lately isn't a new model, but a new workflow. Now that the Claude Code interface is fully available on the mobile app, I’m finding myself working in a new and different way. It used to be that if I had an idea while walking the dog or grabbing coffee, I’d have to hope I remembered the nuance by the time I got back to my keyboard. Or I’d need to open a notes app or voice dictation and just get it down. Now, I just pull out my phone and partner with the LLM right there. I can start ideating, review the architectural approach, or even debug a concept while I'm away from the desk. It’s not about replacing the desktop; it’s about continuity. By the time I actually sit down at my computer, the "blank page" phase is already over. I simply pick up the thread where we left off and keep moving.

The Bottom Line The goal of this toolkit isn't to have the most subscriptions; it's to lower the activation energy required to build things. • Claude/Opus builds the thing. • Gemini plans the thing. • ChatGPT researches the thing. It’s a fragmented workflow, sure, but right now, it’s the most effective way I’ve found to get out of my head and into the work.