Ep 1089 News 5:14 w/ Jessica & Cathy

Hark releases an AI personal assistant with a focus on privacy | TechCrunch

Hark Pro arrives as a privacy-focused personal assistant that can act across a user's digital life. Jessica and Cathy dig into whether visible computer use and task completion can earn trust, or whether the privacy promise breaks down once an assistant needs access to everything.

Embed this episode

Paste this on any site — the player is a self-contained iframe with no cookies or trackers.

<iframe src="https://sandrise.io/exploring-next/embed/1089"
  width="100%" height="180" style="max-width:640px;border:0;border-radius:12px;overflow:hidden"
  title="Exploring Next — Episode 1089 audio player"
  loading="lazy" allow="autoplay" referrerpolicy="strict-origin-when-cross-origin"></iframe>
Embed & API docs →
Script GPT-5.6 terra Voice Cartesia TTS

Transcript

Jessica The hard part isn't getting an assistant to book a flight. It's getting you to hand it the keys to your whole digital life.

Cathy Right, and Hark Pro is asking for a lot of keys. Email, calendar, hard drive, credit cards. Then it uses a computer-use model to do web tasks instead of just drafting a helpful little paragraph about them.

Jessica Which is why this lands now. Everyone is hunting for the personal-assistant interface that feels less like another chat tab and more like a thing that actually clears your day.

Cathy Hark's version is full-screen, with a central chat box, suggested actions, and these mini dashboards they call panels. It can surface an expense approval, an unanswered email, a meeting, or a flight booking before you think to ask.

Jessica That flight example is doing a lot of work, though. We have collectively made that poor flight-booking demo carry the entire future of computing.

Cathy It has been promoted beyond its competence. But Hark's car-registration example is better. An agent noticing a renewal and finishing it through the California D M V site is mundane in exactly the useful way.

Jessica Oh, that's the version I want. Not “look, it knows a fact,” but “that annoying obligation is no longer sitting in the back of my head.”

Cathy Yeah. Hark is at least aiming at substitution, not assistance theater. Its design lead says it may be slower than Muse, but the claim is that it gets to the end without dumping the messy last step back on you.

Jessica Right.

Cathy That is a real product distinction if it holds. A lot of agent demos are basically a relay race where the human gets handed the baton at the one point where the website gets weird, payment is involved, or the form has an ambiguity.

Jessica And Hark makes one smart trust choice: it shows the agent browsing in a small window. You can watch it navigate instead of receiving a mystical “done” message. I care less about seeing every click than knowing I can catch it before it makes a bad assumption.

Cathy Mm-hm.

Jessica There is a tiny bit of life texture in that, honestly. Watching an agent fill out a form is not thrilling, but watching it hesitate at the right moment? Weirdly calming. That's apparently where we are after eleven months of doing this show.

Cathy And Hark enters a crowded race with Muse, Instinct, and Dots all chasing the same basic prize: the assistant people leave open all day. Hark's angle is less “we made a magical brain” and more “we built an interface that can operate the machine.” I find that narrower framing more believable.

Jessica Okay okay.

Cathy But the architecture creates a sharp trade-off. A model trained specifically for computer use can be faster and cheaper on browser actions. Yet a personal assistant also gets vague requests, unfamiliar services, general questions, and all the strange context people never state out loud.

Jessica Exactly, Cathy. That is still the live fight with personal assistants: usefulness needs an intimate profile, while privacy needs boundaries you can actually understand. “Task-related” data sounds reassuring until your email, contacts, purchases, and calendar are all tangled together.

Cathy Yeah.

Jessica Hark says it isn't building an ad business or trying to harvest data, and I don't dismiss that. But privacy becomes a product advantage only when I can tell what stays isolated, what gets remembered, and what the agent can carry from one task into the next.

Cathy Oh no.

Jessica What?

Cathy The custom background changes with your local weather and time. We have given the assistant access to our receipts, our calendar, and our credit card, but it also needs to set the mood like a very expensive aquarium.

Jessica Stop it. Although, yes, that is exactly the kind of polish that makes a product feel alive before it's earned the right to be trusted.

Cathy My concern is less the wallpaper than persistence. If Hark remembers enough to proactively flag a car registration, then wrong assumptions can persist too. We talked about that with proactive agents already: ambiguity becomes state, and state makes the mistakes stick around.

Jessica I think you're right, and this is where Hark has to prove more than a nice surface. The user story works when the assistant takes something off your plate. It fails when checking its work becomes a second, quieter job.

Cathy The visible browser window helps with that, at least. It doesn't solve security or data handling, but it gives the person a way to inspect the path. That's more useful than a confidence score pretending the agent knows when it crossed a line.

Jessica So I’m cautiously excited about the software, not the mysterious twenty twenty-seven device. “Never existed before” is a fun sentence. It is also a sentence that has made a lot of very ordinary hardware sound briefly extraterrestrial.

Cathy Fair. Let the device earn its own episode. For now, I want to see whether Hark can complete ordinary, slightly ugly web chores without quietly turning every user into its operations manager.

Jessica That feels like a good place to leave Hark, Cathy. If the assistant can handle the ugly chores, maybe the keyboard finally gets a smaller role.