A new AI tool launched today, and I've been spending time with it out of curiosity: meet Hark.

I try most notable AI tools usually the week they launch, and watching generative tools evolve still has room to surprise me. This one is organised differently, and that's what I want to note down.

A few observations so far:

- It's one continuous conversation. There's no "new chat" button, and longer work goes into separate Projects.

- It carries out tasks as well as answering. It uses its own browser on its own machine, and it confirms important decisions, like a payment, before acting.

- There's a Home screen of "Panels": widgets you create from any data, just by describing what you want to track.

- Usage resets daily rather than weekly, at least for now, and so far the daily allowance has felt generous.

Visually, it reminds me of Gemini's new Neural Expressive design language, but simplified even further, so it stays as non-invasive as possible. And it does more than I can get from Gemini in a browser tab: the web interface is all you need, and the heavy lifting happens on Hark's side.

The way I see it, Kimi Claw set the direction with its well-known bot mascot, Grok Bot followed, and OpenAI arrived with Dots. Hark takes a different route: no mascot, it just works and feels clean. What stands out so far is that you can hand over a task and genuinely forget about it.

Which brings me to the biggest constraint: trust. The less a tool asks you to manage, the more it matters what it keeps, who can see it, and what it's used for. Right now, opting out of data training with Hark means emailing their team. I'd like to see that become a simple setting soon.

Too early to draw conclusions. I'll keep posting notes as I go, and keep watching how adoption unfolds.

Disclosure: I'm on a free early-access month. No affiliation.