There’s no shortage of new releases in the AI world these days. New models and partnerships emerge weekly it seems. It’s a lot to keep track of and, I suspect, enough noise to force most people to tune out and just stick to what they are using. There must be some psychological phenomenon where too many choices actually yields no choices.
But here’s the thing: I do use multiple tools. Not because I’m a sucker for the shiny and new, but because they’re genuinely better at different things. And “genuinely better” means noticeably, practically, saves-me-real-time better.
Before you go spinning up accounts on fifteen different platforms, ask yourself two questions:
- What do I actually do regularly? Not “what sounds cool” or “what might I need someday.” What tasks eat your hours this week?
- How much time and money do I want to invest in this? Because every tool has a learning curve, and most have a price tag. Your AI stack is a budget line item now, whether you like it or not.
For me, that shakes out to three categories. Yours will be different—and that’s the whole point.
My practical stack
1. General conversation & brainstorming: ChatGPT
When I need to think out loud, work through a problem verbally, or get quick answers while doing something else, ChatGPT’s voice mode is the differentiator. It’s not about the model being “smarter”—it’s about the interaction pattern fitting how I actually work. It’s perfect for my 10 minute car trips to pick up a kid from practice.
2. Images and video: Gemini (via NanoBananaPro)
This one surprised me. I didn’t expect to care much about image generation for work purposes. But NanoBananaPro has been genuinely excellent at producing both professional artifacts and creative stuff with equal precision. Pro tip: meta-prompt it to create a structured JSON prompt for your image request. You can actually use Gemini (or build a gem) to create the prompt and then flip over into “Create Image” mode. The precision jump is meaningful.
3. Coding: Claude
Not even close. Claude is my go-to, and it’s a perfect example of why both the model and the agent matter. The model (Opus 4.5 lately) is strong, but it’s optimized specifically for software engineering workflows when using inside of Claude Code. It feels like a full level up from where things were six months ago.
What I’m NOT doing
Are there incredible AI tools for generating music? Probably. Video editing? Sure. Niche research applications? I assume so.
Do I use any of them? No. Because I don’t make music. I don’t edit video regularly. And I don’t have the bandwidth to evaluate tools for tasks I do twice a year.
This isn’t about finding the best tools. It’s about finding your tools.
Right Tool for the Right Job
We’re in a multi-model world right now. That might change—maybe one platform will genuinely do everything well eventually. But today? Picking one tool for everything means accepting mediocrity somewhere in your workflow.
The good news: you don’t need to learn them all. Learn the ones that match your actual work. Keep your stack small enough that you can stay current when things change (and they will—hi, Gemini 2.5 Pro).
Three to four tools, max. Know them well. Stay open to swapping one out when something legitimately better emerges.
That’s it. I think you’ll find that approach useful!
What’s in your stack? I’m genuinely curious whether other people have landed on similar categories or completely different ones.