Hark Pro Explained: Brett Adcock's Proactive AI Assistant vs Muse, Dots and Grok Bot

On October 6, 2026, Brett Adcock, the founder of the humanoid-robot company Figure, announced Hark Pro with the line "now live on web, iOS, and Android. Try it free." It comes from Hark, a separate company Adcock started in late 2025, and it is the latest entrant in the crowded category of proactive personal AI agents alongside Meta's Muse, OpenAI's Dots and SpaceXAI's Grok Bot.

This post explains what Hark Pro is, what has been reported about pricing, models and funding, how it compares with its rivals, and what to check before handing it your inbox and cards. We worked from Adcock's announcement thread and press coverage and have not tested the product.

TL;DR: the questions people are asking

What Hark Pro does

Hark's pitch is that most assistants are a chat window with shortcuts, while Hark Pro is a place you arrive at to find the day's work already prepared. Reported features:

• A full-screen home with a central chat box, suggested actions and customizable "panels," small dashboards you arrange.
• Proactive suggestions. The agent looks across your connected accounts for work it can take off your plate, such as noticing a child's lunch card balance is low and offering a refill, or noticing a flight is booked but a hotel is not.
• Task execution on the web. It can buy groceries, book cars and flights, pay bills, handle forms such as a DMV registration, approve expenses and answer email. A small window shows the browser as it works, for transparency.
• Handoff. Hark's browser-use system, which we covered in an earlier roundup of Hark's Handoff launch. Reports say it can log into "millions of sites" and manage up to six browsers at once in the cloud.
• An encrypted vault for payment details and passwords. Hark says it cannot access the contents.
• Action buttons for common jobs such as paying a bill or cancelling a service.

Design was led by Abidur Chowdhury, an ex-Apple designer who unveiled the iPhone Air, which explains the emphasis on interface. One user in the replies asked about phone calls: the agent can message and email, they said, but calling was missing and mattered to them. Adcock replied, "We'll get there."

The model claim

Hark says Hark Pro runs on a model trained specifically for computer use, which it says is faster and cheaper for automation than general frontier language models, and that it completes tasks end to end better than competitors such as Muse, Dots and Instinct. The trade-off, which coverage notes, is that a specialized model may have less broad knowledge than a frontier model.

This is a plausible design: for clicking through websites reliably, a model tuned on browsing trajectories can beat a larger generalist, as other computer-use systems have shown. It is also exactly the kind of claim that needs independent testing, because "completes tasks end to end" depends on which tasks, which sites and how failures are counted. Compare it with our look at how agents fail when sites change. A browser agent inherits every website's defenses and layout changes.

Pricing, funding and hardware

Treat the funding line carefully: it appeared in one outlet's report and we have not seen a company filing. The tiering mirrors Meta's Muse, which launched weeks earlier with the same price points, a coincidence that suggests the market is converging on a $20 and $100 structure for proactive agents, a pattern we noted in our five-agent comparison.

How it compares

A reply asked why anyone would switch from Muse, which "does the same thing and more." That is the right question, and the honest answer is that nobody knows yet. Differentiation will come from reliability on real tasks, trust and price, not feature lists. Hark's strongest argument is specialization and design. Its weakest is that it is new, with no track record.

Privacy and risk: what to check

A proactive assistant that can read your email, see your calendar, use your cards and log into sites is powerful and risky in equal measure. Hark says it is "not here to sell you ads and steal your data," positioning itself against Meta and OpenAI, and that its vault is unreadable to the company. Questions worth asking:

1. What does the agent store? Emails, documents and browsing history, for how long, and where.
2. How does the vault work? Is it end-to-end encrypted, who holds the keys and what happens on a device change?
3. What approvals does it require? Can it spend money or send messages without confirmation? Our post on permission lessons from a disputed Grok Bot claim explains why approval gates matter.
4. Which models and vendors see your data? If Hark uses third-party models or services, which ones?
5. What is the liability if it errs? Payments initiated by software raise consumer-protection questions, especially outside the US.
6. Can you revoke access fast? One-click disconnect and a log of actions.
7. Security review. Has an independent firm tested it?

For payments specifically, regulators in Europe require strong customer authentication, which The Next Web cites as a reason availability there is unclear. Expect similar friction in other regulated markets.

What to do now

• If you are curious, try the free tier with a throwaway or low-stakes account. Do not connect banking or primary email on day one.
• If you already use Muse or another agent, run the same three tasks on both, such as a bill reminder, a travel booking and an inbox summary, and compare errors and time.
• If you build agents, note the playbook: a specialized model, a visible browser window for trust, an encrypted vault and designer-led polish.
• If you follow the hardware angle, wait for details. "A product that's never ever existed before," as the design lead put it, is not a specification.

A practical test plan for any proactive agent

Because Hark Pro enters a category with several similar products, a repeatable test is more useful than a launch-day impression. Pick five tasks that reflect your real week: summarize an inbox and draft three replies, find and book a flight within a budget, pay a small recurring bill, cancel a subscription, and fill a government or school form. Run each on two agents, and record whether it finished, how many corrections you made, how long it took and whether it asked for approval at the right moments. Repeat on a different day, since websites change and agents that worked on Monday can fail on Thursday. Keep the stakes low: use a dedicated card with a small limit, a spare email account for the first week, and never allow irreversible actions without confirmation.

Pay attention to failure behavior as much as success. A good agent stops and asks when a page looks different, reports what it could not do and leaves the account in a clean state. A poor one guesses, retries blindly or reports success it did not achieve. That difference matters more over a year than a few points of benchmark accuracy, and it is the difference you only learn by testing.

What this means for what you build or pay

For consumers, the proactive-assistant category now has several credible options priced at the same $20 and $100 points, which should push providers to compete on reliability and privacy. For builders, the interesting signal is that a founder with a robotics company is investing in software agents and then hardware, suggesting the assistant interface is seen as the next platform battle. We will update this post when independent reviews of Hark Pro and clearer terms on privacy and pricing appear.

Related reading

• Conduit Telepathy AI and Hark Handoff (August roundup)
• Dots vs Grok Bot vs Muse vs OpenClaw vs Hermes
• Is Meta's Muse safe to use?
• OpenAI Dots launch
• Grok Bot will route to Claude Opus 5.5, Midjourney and Suno
• Meta Muse daily users plateau
• Agent teams and the permission bottleneck
• Gumloop Agent Browsers and AgentID

Primary: Brett Adcock's announcement on X (October 6, 2026) · TechCrunch and The Next Web coverage of the launch · hark.com

Details are accurate as of October 7, 2026 and come from the announcement and press coverage. We have not tested Hark Pro. Pricing, funding figures and availability are as reported by outlets and may differ from the company's own materials; claims about speed, accuracy and privacy are the company's.