AI Tools I Actually Use

· 4 min read
AI tools productivity LLM

I build software that incorporates AI, which means I spend a lot of time thinking about what these tools can and can’t do. But I’m also a user. Here’s what I actually use day-to-day, stripped of any evangelism or hand-waving about the future.

Claude Desktop: My Main Interface

Claude Desktop is my general-purpose AI interface. I use it for most things that aren’t code, like drafting, research, architecture discussions, etc. It’s not perfect, but it’s become a pretty central part of my workflow.

What makes it particularly useful for me is the Model Context Protocol (MCP) integration. I’ve built three custom MCP servers that connect Claude to my actual work:

  • Evergreen - My personal CRM. Tracks relationships, interactions, and follow-ups. Claude can look up contacts, log meetings, create action items, and even enrich profiles with web research. Makes relationship management feel less like database maintenance.

  • Lexicon - A prompt library that turns into slash commands. I save prompts I use repeatedly (like “run linter and fix issues” or “audit for tech debt”), and they become available as options in the desktop app. Saves me from retyping the same instructions.

  • ContextSwitch - A kanban board that lives in my git repos as SQLite databases. Claude can read my task list, create new tasks, mark things done, and track dependencies. Lets me switch between different AI tools without losing project context.

These aren’t general-purpose tools. They’re personal software I built for specific workflow problems. But having Claude connected to them transforms it from a general chatbot into something that knows what I’m working on and who I’m working with. The difference is substantial.

Granola: Meeting Notes That Don’t Suck

I’ve tried a bunch of meeting note tools, and Granola is the one that stuck. It records meetings, transcribes them, and generates useful notes without making me think about it.

The key thing Granola gets right is staying out of my way. I don’t have to remember to start it, configure settings every time, or wrestle with bad transcriptions. It just works, and the output is actually useful.

NotebookLM: Quick Topic Overviews

NotebookLM has this feature where it generates a podcast-style conversation about whatever you feed it. I use it when I need to get up to speed on a new topic quickly.

The caveat: it rarely goes deep on technical details. It’s more like listening to two people chat about a subject at a high level. But for that initial “what is this thing and why should I care” phase, it’s surprisingly effective. Better than skimming a dozen blog posts to get oriented.

Whisperflow: Occasional Voice Input

I have Whisperflow installed for voice-to-text, but I don’t use it nearly as much as I thought I would. Maybe 5% of the time, when I need to brain dump a bunch of thoughts into a prompt or document.

Turns out I still prefer typing for most things. I think it’s more a matter of habit. Old dog and new tricks or whatever. Voice is good for capturing raw ideas quickly, but that’s a pretty narrow use case for me.

I would imagine I’m on the wrong side of this trend and things like voice interfaces continue to grow, but it still feels a bit unnatural to me.

Google Sheets + Gemini: Surprisingly Handy

The Gemini integration in Google Sheets is one of those things I didn’t expect to use much but occasionally find myself grateful for. Need to clean up messy data, extract patterns, or generate formulas for weird transformations? It handles a lot of the tedious spreadsheet work.

Not something I use daily, but when I need it, I really need it.

What I Don’t Use

AI browsers: The security and privacy implications make me uneasy. Having an AI layer with access to all my browsing data and credentials feels like an unnecessary risk for whatever convenience it might provide. It just feels like a disaster waiting to happen.

Personal advice or topics: I don’t talk to any of these models about personal matters or ask for life advice. My interactions are strictly professional code, writing, research, work tasks. I’m not interested in having an AI as a therapist, friend, or life coach.

The Pattern

Looking at this list, there’s a clear pattern: I use AI tools that augment specific tasks I’m already doing, not ones that try to replace my judgment or become some kind of general-purpose assistant for my life.

The tools that work are the ones that:

  • Solve a concrete problem
  • Stay out of my way
  • Don’t require me to change how I work dramatically
  • Respect reasonable boundaries around privacy and security

That’s where the actual utility is for me. Not in having an AI friend or letting algorithms make decisions, but in getting tedious work done faster so I can focus on the parts that actually require thinking.