This website uses cookies

Read our Privacy policy and Terms of use for more information.

The tools people name and the tools people use are two different lists.

Every week somebody asks me what AI tools they should be using, and for months I've been answering off the top of my head. So I stopped guessing and went and counted.

I scanned 106 meeting transcripts from the last 3 weeks.. client sessions, internal syncs, events at the house, a couple of workshops. For each tool I counted how many separate rooms it came up in, not how many times it got said.

Here's the list organized by type. Fair warning that this is my calendar, so it leans toward the work we do and the people I sit with. By no means meant to reflect AI usage everywhere 😅

1. The everyday assistant (the obvious ones..)

  • Claude. 52 of 106 meetings, more than half of everything on my calendar. It's what we run the company on, and not in the "we have a subscription" sense.. it's where work starts for people who aren't technical.

  • ChatGPT. 18 meetings, almost always as the reference point rather than the tool. It's how people describe where they started, or what they compare against. I can count on one hand the times somebody described doing real work in it.

  • Gemini. 9 meetings, mostly inside lists of model names. The one real use I can point at is generating images and video, where we hand work to it on purpose.

  • Grok. 7 meetings. We're watching it, not running it. Grok Bot came up in a meeting this month as a way to stand up something that behaves like an employee in an afternoon, and that got people's attention in the room.

A few real examples of how I’m using AI as an assistant..

  • The bookkeeping runs itself. I have a skill that does my reconciliations, so I don't touch QuickBooks for them anymore. I didn't write those by hand.. I used Claude to draft/refine/run the workflow.

  • Pairing 60 people before an event. Everyone coming to an event at the CLT Startup House fills out a short survey about what they're working on. I hand that over, ask it to pair people on real overlap, and send everyone a "you should meet this person" note before they arrive. That used to be an evening of work.

  • Taking inventory nobody wrote down. During a transition I had it inventory every domain, account and access we had scattered across a dozen places. Boring, and the kind of thing that never gets done otherwise.

I don't do the reconciliations anymore.

2. Building software

  • Claude Code. 23 meetings, and the deepest real adoption in the set. The number that matters is who's in it.. our COO opens pull requests, and people who've never written a line of code ship real work with it.

  • Codex. 8 meetings. We don't build with it, we point it at work another model already wrote. A second model coming in cold catches things the author is structurally blind to.

  • CodeRabbit. Runs on every pull request and looks for bugs/issues in the code. Boring, and it earns its keep.

  • Lovable. 8 meetings. Real but casual, and firmly the prototyping tier.. somebody described throwing up an ugly, horrible prototype in a day just to see whether an idea had legs. That's the right use.

  • Replit. 7 meetings, all of it client-side. One client's team is being taught to build on it, which is a different job than us building for them.

  • Cursor. 3 meetings, every one of them inside a list of tool names somebody was reciting. Zero meetings where anybody described doing work in it that week. Loud in the discourse, quiet in my calendar.

  • Devin. 5 meetings and can be summed up in this one quote: "I've stopped running Devin."

3. Meetings, notes and voice

  • Fathom. 18 meetings. Records everything, and it's where this whole count came from. My first scan crowned it our most-used tool by a mile, until I noticed every transcript carries a fathom.video link in its header.

  • Otter. 2 meetings, which undersells it badly. It runs constantly and nobody ever discusses it, which is roughly the highest compliment infrastructure gets.

  • Wispr Flow. Voice as the actual interface, not a novelty. On the drive home I open a session and talk at it, and the agents execute against what I said. A logo project this month started that way.

  • MCP. 9 meetings. Less a tool than the plumbing connecting the assistant to everything else. It's become the word people reach for when they mean integration.

4. What the mid-market already pays for and doesn't open

  • Copilot. 12 meetings, and most of those mentions aren't what Microsoft is hoping for. In the rooms I'm in it's the thing a company already bought, rolled out to a few people, and left there. That conversation usually comes down to training people properly or admitting nobody's going to open it.

  • n8n. Client-side automation at one account, currently being switched off as part of a transition. Fine tool, wrong layer for us.

The gap between what a company licenses and what its people open is the most common thing I walk into, and it's rarely a tool problem.

5. The 8 I vetted in the last 2 weeks

A list of what I use is only half an answer. Over the last 2 weeks I did full teardowns on 8 agent tools and platforms.. cloned them, read the install scripts, checked the telemetry, read the terms. Where each landed:

  • pi. A coding agent that runs non-Anthropic models. Adopt, narrowly. Good for work I deliberately want off Claude, though a Claude subscription doesn't cover a third-party harness.. that's metered usage on top.

  • open-slide. Decks as React components an agent can actually edit. Adopt, narrowly. Internal and browser-presented decks only, not deliverables.

  • PipesHub. Open-source Glean, so search across all your company's documents. Adapt. Useful, and its telemetry ships to a third-party domain by default, so read that before anything sensitive goes near it.

  • Pipecat. A framework for real-time voice agents. Defer. Good software, and I don't need it.

  • DeepSeek harness. DeepSeek's own agent harness. No, but steal from it. Three patterns went into ours.

  • Amp orbs. Isolated sandboxes so parallel agents stop tripping over each other. Don't migrate. It isolates compute; our bottleneck is a shared hosted database, so it solves a problem we don't have.

  • backpass. Mines your old agent sessions for reusable patterns. No, on fit. Good engineering; our transcript corpus just isn't structured the way it needs.

  • DeepAPI. A billing wrapper over scrapers, search and hosted models. No for us.

Wrapping up

It’s easy to look at a list of tools like the ones I outlined above and think “Oh crap, I’m not using these and have never heard of them.” That’s not the takeaway.

Instead it should prompt exploration and curiosity to experiment/test (safely) and see if there are nuggets you can glean.

Also, important to remember that the best tools are the ones you have with you and actually use 😉

What I'm reading

Onward & upward 🤘

Drew

P.s. If we haven't met yet, hello. I'm Drew Burdick, Founder and Managing Partner at StealthX. We work with brands to design and build great customer experiences that win. I share ideas weekly through this newsletter and over on the Building Great Experiences podcast. Have a question? Feel free to contact us. I'd love to hear from you.