The AI personal assistant you're looking for doesn't exist (yet)

Hey!

A friend reached out last week. They’re running a startup and wanted to know if I could recommend an AI personal assistant. Something that handles the inbox, manages the calendar, keeps the chaos organized. Reasonable ask. The kind of thing I should have an easy answer for.

I didn’t.

I build customized solutions for clients all the time. But this friend wasn't asking for a custom build. They wanted something they could just use, something off-the-shelf that actually works. And I realized I didn't have a clean answer for that.

So I went back and actually looked at what’s out there right now.

The options, honestly

​Lindy was the first stop. And honestly, it impressed me. The way it learns your preferences, adapts to your communication style, stays in the loop across conversations. It’s genuinely well-designed for what it is. But then I saw the pricing. €50 a month. For a solo founder or a small startup watching every line of their P&L, that’s not a trivial ask. Especially when “what it does” is still a pretty specific slice of what a real personal assistant handles. Though admittedly, they are expanding into other areas like Lead Generation (currently testing it).

Then there’s Cowork, Anthropic’s Claude desktop feature, which I’ve been using a lot lately. It sits in a different category from Lindy. You’re not buying a product; you’re building your own setup with Claude as the engine. I’ve replaced several paid subscriptions with skills I built and scheduled myself. No per-task fee, no lock-in. The trade-off is that it asks more of you upfront. And fair warning: skills in Cowork are not the same as skills in Claude.ai, which are not the same as skills in Claude Code. The terminology is a mess right now and the ecosystem is still finding its shape. Plugins help with sharing and extending what you build, but this is still “build your own” territory.

Next I looked at hosted OpenClaw options. KiloClaw stood out because it removes a lot of the technical barrier to entry, which is genuinely useful. But here’s the honest read: even with that, it’s still not for everyone. If you’re not particularly technical, and especially if you’re not sure how to protect yourself when something goes sideways with your data or your integrations, there’s still a lot of open uncertainty for actual business use. It’s promising. It’s not solved.

And then there’s building your own from scratch. Which is what I keep gravitating toward for anything with very specific requirements. App Scripts for the lightweight stuff. Mastra agents when you need something with more reasoning. The upside is full control and zero monthly fees. The downside is that it takes time to build, time to maintain, and time to debug when Gmail changes something quietly and your triage logic starts misfiring.

What I told my friend

None of these options is “the answer.” What actually exists right now is a patchwork. Lindy if you can justify the cost and want something that just works. Cowork if you’re willing to build but want to own what you build. KiloClaw if you want more control than Lindy but aren’t ready to build from scratch. And your own custom solution if you need it to fit exactly how you work.

And a healthy dose of skepticism toward anyone who tells you their entire business “runs on AI.”

Because that claim is almost never wrong. It’s just incomplete. What they usually mean is: certain parts of their business run on certain tools. The inbox triage maybe. The calendar scheduling maybe. And then everything that requires real judgment, the things where getting it wrong actually costs something, a human is still in the loop. Which is fine. That’s actually the smart call. But it’s not the same as “the business runs on AI.”

The gap between a working setup and something that runs reliably without daily babysitting is where most of these projects quietly stall. I wrote about what it actually takes to get AI agents to that point this week if you want the full picture. And if you’re trying to figure out where the human should stay in the loop versus where you can safely let the AI run, this piece on the augment vs. automate decision is the framework I keep coming back to.

What this means for you

The tools that save you the most money are usually the ones that ask the most from you upfront. The tools that ask nothing upfront charge you on the back end: in money, in lock-in, or in the moment they don’t do what you assumed they would.

I’ve been thinking about this a lot with Cowork specifically. Some of the workflows I now run there, I used to pay monthly subscriptions for. The cost is zero. But I spent real time setting them up, and I’ll spend real time when something breaks. That’s the trade. It’s a good trade for me. It won’t be for everyone.

Before evaluating any AI tool for your workflow, the most useful question isn’t “what can this do?” It’s “what happens when it gets it wrong, and who catches it?” That question will tell you faster than any feature list whether something actually fits how you work.

I’m going deeper on Cowork soon. If you’re already experimenting with it, hit reply and tell me what you’re running. Curious what problems people are actually solving with it.

P.S.: If this newsletter has been useful to you, I'd love a short testimonial. It helps more than you'd think. Takes 30 seconds.