The AI Tools I Actually Use Every Day (2026 Edition)

Why This List Is Different

Every few months someone publishes "The 47 AI Tools You Need in 2026" and it's always the same: a sponsored carousel of logos, affiliate links dressed up as opinions, and tools the writer tried once for twenty minutes. You can feel the lack of use in every sentence.

This isn't that. No sponsors, no affiliate links, no tools I haven't used for at least a month straight. Just an honest account of what's in my actual workflow and what each tool replaced.

I work across writing, music, code, and AI research. My stack reflects that breadth. Your stack will look different, and that's fine. The goal isn't to copy mine. The goal is to show you what it looks like when someone actually commits to a tool long enough to know whether it's earning its spot.

Writing and Thinking

Claude (Anthropic)

I use Claude for everything that requires extended reasoning: working through arguments, stress-testing ideas, drafting copy, and having the kind of back-and-forth that sharpens a thought before it becomes a post. It handles nuance better than any other model I've used, and it doesn't flatten your voice the way some tools do.

What I don't use it for: generating copy wholesale and calling it mine. That's not writing. I use it the way I'd use a sharp editor. It pushes back, catches inconsistencies, and asks the question I was hoping to avoid.

Obsidian

All my notes, every idea, every draft outline starts in Obsidian. Local files, plain markdown, no lock-in. I've been using it for three years and I've never once worried about losing my work or a company going under and taking my thoughts with it.

The graph view is mostly decoration. The value is the discipline of writing things down in one place, consistently, in a format you'll be able to read in ten years.

"The best writing tool is the one you actually open. Everything else is just preference dressed up as principle."

Building

Cursor

Cursor is the only coding environment I've used where AI assistance feels like it's actually in the loop with what I'm building rather than just generating plausible-looking code to copy-paste. The tab completion is fast enough that it doesn't break flow, and the ability to talk to it in context saves hours on anything involving unfamiliar codebases.

The thing that kept me on it: it got better at saying "I don't know" instead of making something up. That matters. Confidently wrong suggestions are more expensive than honest uncertainty.

Claude Code

For multi-file changes, refactors, and anything where I need to stay in the terminal, Claude Code handles what Cursor doesn't. The two tools overlap, but they're not redundant. Cursor is for flow state. Claude Code is for surgical work where I want full control and a clear audit trail of what changed and why.

Research and Search

Perplexity

Perplexity replaced Google for almost everything except things I know Google indexes better (local services, very recent news). The cited sources are what make it worth using. You can see exactly where a claim comes from, which means you can spot when the answer is wrong before you act on it.

The deeper reason I use it: it changed how I search. I stopped asking "how do I X" and started asking "what's actually true about X and what do I not know yet." That's a better question, and Perplexity rewards asking it.

What Got Cut and Why

Tools leave the stack for one of three reasons: they stopped working, something better replaced them, or they were solving a problem I no longer have.

• ChatGPT. Replaced by Claude for reasoning and drafting. I still open it occasionally to check whether outputs differ on a specific task. Usually they don't differ enough to matter.
• Notion. Replaced by Obsidian. Notion is a great tool with a fundamental problem: it wants to be everything. When a tool wants to be everything, it becomes optimized for nothing you specifically need.
• Several "AI writing assistants." Grammarly, Jasper, and a handful of others I tried for more than a week. They optimized for sounding confident and polished. I'd rather sound like myself.
• Every "AI productivity dashboard." Tools that aggregate other tools don't save time. They add a layer of indirection before you can do the actual work.

"Cutting tools feels like losing capability. It isn't. A smaller stack you actually use beats a large stack you're always configuring."

How to Pick Your Stack

One question filters out most tools worth your time: does this tool make a thing I already do faster, or does it ask me to do a new thing to benefit from it?

Tools that make existing things faster compound. Tools that require new behaviors usually don't make it past the first week.

Secondary filters, in order:

1. Would I pay for it with my own money? If your employer pays and you'd cancel the day you left, it's not actually a part of your workflow.
2. Does it get out of the way? The best tools are nearly invisible. If you're aware you're using a tool, it's still competing with the work rather than serving it.
3. Can you export your data? If the answer is "with difficulty" or "not really," your investment in the tool is actually an investment in their platform. Build your stack on tools that don't hold your work hostage.

Build slow, cut fast, stay ruthless. Your stack should earn its cognitive rent every month. If it isn't, it's a liability pretending to be an asset.