What will it take to launch the first must-have AI consumer product? Maybe $700 million.

At least according to Hark, an AI lab building models and hardware for an AI personal assistant, which said on Thursday that it had raised that much in a Series A round that values it at $6 billion post-money.

The mega round was led by Parkway Venture Capital, and included Align Ventures, AMD Ventures, ARK Invest, Brookfield, Greycroft, Intel Capital, Prime Movers Lab, Qualcomm Ventures, Salesforce Ventures, and Tamarack

Global. (Phew!)

Perhaps what’s most notable about the fundraise is how little Hark has revealed about what it is building. Founder and CEO, Brett Adcock, also the entrepreneur behind robotics company Figure.AI and electric aircraft builder Archer, launched Hark in late 2025 with $100 million of his own money to develop an agentic AI system that serves as a universal interface with the digital world.

Hark expects to release its first multi-modal models this summer, which it says will power a personal AI platform that works with existing products and services. The company expects to follow that with hardware devices built specifically for those systems.

The fresh cash will be spent on recruiting top talent for hardware, product design and AI research, and on securing compute and components. The company currently has 70 employees, and runs a data center with Nvidia B200 GPUs.

Abidur Chowdhury (pictured above in a promo video), a former Apple product executive, is Hark’s director of design. He declined to reveal new details of what he’s working on when TechCrunch peppered him with questions this week, but said investors were impressed by a series of demos from his team.

“I haven’t seen anything that feels like something that will really help like the normal person,” Chowdhury said, speaking of the AI products on the market. “People are really building things to help people make software, and it’s working, and it’s really impactful, but we haven’t really seen that for the normal person yet.”

He noted that while Anthropic is prioritizing coding tools and OpenAI is moving in the same direction ahead of its IPO, few companies are focused solely on building interfaces and native hardware the way Hark is. “With this focus, with this great team that we have, and this round that we’ve raised, I think we can make something really special in this space,” Chowdhury said.

Still, there are more questions than answers. One challenge will be providing the context of a customer’s life to an AI assistant without making the people around the user uncomfortable or violating their privacy. Wearables like Meta’s existing glasses or the forthcoming Android spectacles don’t seem to have solved this problem. When asked how he might square this particular circle, Chowdhury only smiled.

“Sounds like that would make a great product.”

One of the most promising introductions at Google’s I/O developer conference on Tuesday was a new way for consumers to use the web: AI agents. Unfortunately, it was also the most confusing.

Google took the wraps off information agents, a reinvention of the aging Google Alerts service, now infused with AI. These AI agents are designed to operate in the background, 24/7, helping users stay up to date on topics they’re interested in, like market trends, price tracking, or inclement weather warnings.

Information agents Image Credits:Google

Then there is Google Spark, a “personal” AI agent that can help you navigate your digital life by integrating with Google products, like Gmail, Google Docs and Google Workspace. The company says the assistant can handle everyday tasks like surfacing themes from newsletters, organizing your home inventory and keeping track of what needs restocking, or helping you plan and manage a group trip with friends.

Or, as Google showed off in a very engineering-minded example, you could use it to organize a neighborhood block party — as if that would require any management beyond a group chat or some emails.

Gemini Spark Image Credits:Google

There’s also a name for how you track notifications from Spark: Android Halo. (Why an Android feature needs its own brand is beyond me, but a good guess is that Google’s internal product teams are fairly competitive and want to highlight their own work, even at the risk of confusing users.)

Image Credits:Google

Next, Gemini’s app is getting an AI agent that can compile a personalized digest from your Gmail inbox, calendar, and tasks, and provide an update called Daily Brief.

Image Credits:Google

Many of these products have not yet shipped, or at least won’t be available to the wider public right away. Instead, Google is targeting its heavier users for now: the “AI-pilled” subscribers of its new, only $100-per-month Gemini Ultra plan.

Google Pro and Ultra subscribers in the U.S. will get to use Information agents starting this summer, and Spark will be available to Ultra subscribers “soon.” Halo will ship to Android users “later this year.” Daily Brief is rolling out in the U.S. to Ultra, Pro, and Plus subscribers.

Image Credits:Google

As a result of all these launches, we’ll soon have so many entry points for using AI agents that it may be overwhelming as to where to start. (Did I forget to mention the increasingly agentic Chrome web browser, too? Google showed off how you could talk to Chrome while shopping for cars online to configure the various options and trim levels you can afford without tapping on a keyboard and clicking around. Yay…I guess?)

In a press briefing ahead of I/O, Google said it intends to bring its agentic features, including Spark, to free users “when the time is right.” But for the time being, the company’s more interested in iterating with a group of people, like the Ultra subscribers, who will push the limits of what Spark and AI agents can do.

Image Credits:Google

In the meantime, Google is furthering the divide between those who have already bought into (literally!) the promise of AI, and the average consumer using Google’s free tools, who’s likely distanced from the real-world improvements AI offers, like agentic coding or AI-enabled computer use.

Instead, consumers today largely think of AI as chatbots replacing traditional Google searches. They think of AI photo and video models not as impressive creative leaps, but as tools for making “AI slop” that now clutters their social feeds, and result in unwanted data centers being built in their backyards.

Google didn’t help its reputation on this front during the event, flashing goofy AI imagery between each presenter. It also played a corny AI-generated animation featuring Cinnamon Toast Crunch-esque talking Tensor chips. And in its Android glasses demo, Google showed how the devices — which will later support photo-taking — could use AI to transform photos users take into something else.

Image Credits:Google

This demo involved the presenter taking a picture of their view of the audience, which was modified to have a blimp floating overhead, and then sent to their Android Watch. Okay, neat, but is it worth someone’s home being torn down via eminent domain to build new power lines for a data center?

People will need more than clever party tricks to accept such drastic societal changes.

Image Credits:Google

In previous years, Google introduced new consumer electronics devices, like Pixel phones and Nest Hubs, alongside new Android features, like that restaurant-and-salon booking service that blew people away in 2018. Those pieces of technology were framed as attempts to smooth over some of life’s everyday hassles.

Now, the tech giant is showcasing its new models (but not Gemini Pro 3.5, which wasn’t ready yet) alongside its developer platforms, and largely forgetting about who it’s building all this for: Regular folks. People who don’t want to think about whether it’s called Gemini or Spark or Halo or information agents, or where you go to use it.

These people have real problems they want to solve. They struggle to pay bills and rent, or buy gas or groceries, as they try to find work in the face of AI recruiting systems that reject their resumes over small technical details. They are people who are trying to balance stressful lives that have, of late, come to bear technology’s advances as burdens, especially with social media devouring screen time, addicting children, and turning social connection tools into a big, online shopping mall.

Instead of tools to solve problems, the average tech-savvy consumer watching this year’s Google I/O saw a tech giant putting more AI into everything they use — from Docs and email inboxes to glasses and even Search, which is now more of an AI-first experience.

If Google had tapped real consumer sentiment, it could have noted that AI agents would lower screen time usage. That is, instead of spending time researching, organizing, tracking, and monitoring information and news, agents could take over those daily tasks so users could go offline and live their real lives away from a computer.

Gemini Spark Image Credits:Google

That’s a message that could resonate with consumers, particularly young people, who are today embracing nostalgic retro tech, adopting “old people” hobbies and crafts to de-stress, and rediscovering the power of real-life connections by ditching dating apps for in-person events and experiences.

In short, Google failed to sell just how cool AI agents are by not demonstrating any problems agents solve for everyday users, and keeping these tools paywalled, limiting their reach.

Meanwhile, messaging-first AI startups like Poke, Poppy, RPLY, and Wingman are presenting themselves as a way to interact more naturally with AI agents via a feature everyone uses daily: text messaging.

Will you ever be able to message Spark? Reps at Google I/O vaguely said it will happen at some point in the future.

This is such a different strategy from Google’s early days, when it introduced revolutionary products like Gmail, a free email service that vastly improved on existing options, or Google Search itself, which freely organized the early web and made it more accessible to everyone.

Google I/O could have been a breakout moment when AI agents became available to everyone via a simple, free consumer product (with one brand name!). This product may even have people clamoring for the way they used to beg for Gmail invites. Instead, Google’s new AI agents — tools that can work for us and meet our personalized needs — remain largely out of reach for most.

Stellantis, the automaker behind the Jeep and Ram brands, has tapped self-driving startup Wayve to bring hands-free driving to its vehicles in 2028.

The companies announced the deal Thursday during Stellantis’ investor day at its North American headquarters in Michigan.

This is the second automaker deal for the buzzy U.K.-based startup and comes on the heels of a $1.2 billion Series D funding round that attracted deep-pocketed strategic investors, including Nissan and Stellantis, and returning backers Microsoft, Nvidia, and Uber.

Wayve didn’t disclose the contractual value of the partnership or provide details on which Stellantis vehicles will get the self-driving software “brains” developed by Wayve, but according to Wayve CEO Alex Kendall, this is a commercial contract to supply Stellantis with tech at scale. The companies are targeting the North American market first, which helps narrow the field of Stellantis’ 14 brands, which also include Chrysler and Dodge.

“One of the amazing things about Stellantis is the global, massive scale they operate at, and the diversity of products they offer,” Kendall told TechCrunch describing the opportunity for his startup. “It’s one of the reasons why it’s such a good match because our AI is so adaptable; we can generalize to the variety of products that they offer, and means that because of the diversity of sizes, shapes of vehicles, different driving styles, different geographies they run in our AI is built to scale across them all.”

By 2028, there could be more vehicles to choose from. Stellantis announced Thursday that it plans to expand its market coverage in North America by launching 11 new vehicles by 2030 as part of its $70 billion turnaround plan.

Seven of those vehicles will be priced under $40,000, and two under $30,000, Stellantis said.

It’s not clear if Wayve’s tech will show up in those lower-cost cars and SUVs. Although, if one took Wayve’s efficiency pitch to heart, it might seem plausible.

Wayve has developed a self-driving system that isn’t tied to particular sensors, chips, or high-definition maps, which cost-sensitive automakers like Nissan — and now Stellantis — have found appealing. Instead, Wayve’s software uses an end-to-end neural network that only uses data — captured from whatever sensors are on the vehicle — to direct and teach the vehicle how to drive. Wayve’s software can also run on whatever chip its OEM (original equipment manufacturer) partners already have in their vehicles.

Wayve’s technology supports two products that the company is marketing to automakers and tech companies — a hands-off assisted driving system that’s comparable to Tesla Full Self-Driving (Supervised) and eventually a driverless system designed for robotaxis or even passenger vehicles.

Stellantis will use the hands-off, eyes-on system, a prototype of which was developed for the automaker in just two months, Kendall said. He noted that within a couple of weeks engineers had the vehicle — using the AI-based system — up and driving.

“I think that what we’ve been able to show is that we’ve been able to build a version of FSD that’s built on an AI model that is truly set up to generalize,” Kendall said when asked about how Wayve compares to Tesla’s system. “It’s capable of generalizing across different compute stacks, different sensors, different vehicles, shapes, and sizes.”

When the founders of a mental health app for men called Mental saw that one feature — AI interactive audio — was resonating wildly with their users, they knew they were onto something.

And so the idea for a new, and hopefully safer, kind of AI therapy app was born, which they called The Path, co-founder and CEO Anson Whitmer tells TechCrunch.

Then famed author and motivational speaker Tony Robbins grew so enamored with this startup; he scooched in as a co-founder.

The Path has now raised $14.3 million in seed funding led by Prime Movers Lab (where Robbins is a partner), with participation from speed skater Apolo Anton Ohno, boxer Deontay Wilder, and Designer Fund.

After Prime Movers invested, Robbins began chatting with Whitmer and co-founder Tyler Sheaffer on small stuff like branding, but as his enthusiasm and ideas for the app grew, they offered to bring him in as a co-founder. The author has since helped shape The Path into a therapy-plus-coaching app that taps into Robbins’ popular self-improvements methods.

Whitmer, formerly an early employee at meditation app Calm alongside Sheaffer, says his pursuit of mental health tech was born out of tragic experiences: When he was 19, a beloved uncle committed suicide.

That inspired Whitmar to get a PhD in psychology, and he planned to go into research after graduation. But while he was in college, a cousin left a voicemail. “I didn’t realize until it was too late. It was also a call for help, and he killed himself,” Whitmar recalls.

That spurred a change of course towards work that could bring science’s findings to the masses. Working at Calm was a natural first step, as the research on how meditation improves mental health is solid. Still, after working at Calm until 2021, Whitmar felt he could do more.

“Even though we did have a big impact, it’s not really a big enough impact,” he said. “The issue is, people’s problems are just too idiosyncratic. They’re too personal. They’re unique.”

Plus, everyone will never have access to individual therapy or coaching. There just aren’t enough therapists in the world for that.

The Path, co-founder and CEO Anson Whitmer Image Credits:The Path

Whitmar sees LLMs and AI as the bridge spanning that gap. “What’s exciting and game-changing is that, for the first time in my career, I’ve seen that there’s actually this possibility for every single person to have the personalized sort of access and care that they need to really get the help,” he said.

In fact, such a thing is already starting to happen. OpenAI has said that at least 900 people use ChatGPT for mental health-related queries every week.

However, the problem with using consumer chatbots for mental health is that they are “optimized for engagement,” Whitmer says, and that is the opposite of what therapy and coaching should do.

Consumer chatbots try to solve problems quickly for users, and also engage in “reinforcement” of ideas, to keep users coming back for more. “But therapy/coaching doesn’t work that way. You’re trying to understand the problem deeply,” he said. The idea is to dig out assumptions and then help the person discover their own solutions.

Whitmar says The Path’s AI is trained “to set up structure, so that later on, you can get to a place where there is resolution,” but from a place of understanding.

To that end, Whitmar says the startup’s specially trained AI model has scored a 95 on the mental health safety AI benchmark, Vera-MH. This compares to a top score of 65 for the consumer bots.

“It’s meant to challenge you. It’s not just meant to agree with you,” he says. In fact, he says the app’s model is post-trained from open source models, so it doesn’t use the major consumer LLMs at all, meaning it is not simply a wrapper over them.

The Path, which lets users choose from 11 virtual AI therapists and customize their preferences for directness and other details, is currently free as it gains users. Eventually, the startup plans to charge $40 a month.

In a media landscape dominated by Baby Shark and Skibidi Toilet, one startup is reimagining children’s media by focusing on well-being, not watch time.

Maka Kids is building a streaming app for children ages zero to six featuring content designed for healthy development. The startup has now raised $3 million in seed funding to scale its platform, and is currently accepting waitlist sign-ups.

Unlike traditional streaming platforms, Maka Kids doesn’t have recommendation algorithms, ads, or auto-play. Instead, it is designed to offer a predictable experience that supports learning, creativity, and emotional growth.

Maka Kids was founded by Isabel Sheinman and Tanyella Leta, who previously founded Nabu, a non-profit venture that brought children’s books to more than 15 million children across 26 countries.

Sheinman and Leta were introduced at a dinner back in 2013 through a mutual friend and immediately hit it off, the pair told TechCrunch in an email. They said they initially over the fact that they both came from families of educators and entrepreneurs, an experience that first inspired Nabu and later fueled their passion for Maka Kids.

They began dreaming up the concept of Maka Kids after discussions with their friends, families, and customers at Nabu. They heard from parents who felt increasingly anxious about the effects of screen time on their children. Building on those concerns, the duo conducted hundreds of user interviews, which ultimately shaped their solution: a children’s streaming app designed with well-being at its core.

Maka Kids founders Tanyella Leta and Isabel Sheinman Image Credits:Maka Kids

“We were seeing parents get completely overwhelmed trying to weigh decisions about what was unsafe, what was good, and understand why their kid was melting down every time screen time ended,” Sheinman said. “At the same time, we watched the children’s media ecosystem get louder, faster, more algorithmically driven. Looking at this problem, we felt uniquely positioned to deliver the relief that parents craved.”

All of the content on Maka Kids is evaluated using Maka Imprint, the startup’s patent-pending developmental framework created through two years of R&D in collaboration with researchers at the Yale Child Study Center. The framework maps seven core domains of early childhood development across more than 650 developmental indicators, including language, creativity, emotional skills, and growth mindset.

Maka Kids licenses content directly from IP holders and individual creators. The startup is also partnering directly with studios and animators to produce original content.

Every show on the platform goes through an analysis of pacing, stimulation levels, color contrast, and narrative structure. Its catalog features slower-paced, lower-stimulation content with genuine narrative arcs and stories from around the world.

The duo believes an important factor often missing from the screen time debate for kids is how much the right story, delivered at the right moment, can positively support a young child.

“Stories can support language development, emotional regulation, curiosity, and give kids a sense of how wide the world is,” Leta said. “Children’s media at its best is one of the most powerful developmental tools families have, when it’s designed with this intention. Most of the platforms children watch on today were designed for adult audiences, with a kids experience crudely bolted on as an afterthought. The incentive for the majority of kids’ streaming platforms is watch time, not well-being.”

Image Credits:Maka Kids

When parents create a profile for their child, they can select channels focused on a variety of topics, such as kindness, STEM, emotional regulation, or movement, and then set preferred session lengths. From there, Maka Kids delivers curated, developmentally vetted content tailored to those selections.

The session then ends naturally, with wind-down cues from characters to help children calmly transition away from screen time without a meltdown.

Maka Kids is running a private beta on iOS this summer and plans to launch publicly this fall on iPhone and iPad, with casting support via AirPlay. Maka Kids says it already has thousands of families on its waitlist.

As for the startup’s business model, the app will operate on a subscription model, priced at $11.99 per month, with a discounted annual option.

In terms of the new funding, the startup plans to use it to grow its catalog of vetted shows. The round was led by Michigan Rise, with participation from Union Heritage Ventures, Flybridge, Also Capital, Detroit Venture Partners, Song United, Invest Detroit, Ann Arbor Spark Capital, and Segal Ventures, as well as angel investors.

“Longer term, our vision is to become the trust layer for every digital experience children have,” Sheinman said. “Embedded into games, edtech products, and shows, Maka Imprint can help developers align their products to what is actually good for kids and families. The kids category deserves a trusted industry standard, and that’s what we are building.”

Flipper Devices, maker of the Flipper Zero hacking device, today announced a new gadget called Flipper One that has multiple network connectivity chops and can act as a Linux PC (is this the year?). The company has sold over a million Flipper Zero units and has generated over $150 million in sales. However, the new device is not a successor as it operates on a different layer than the Flipper Zero, the company said.

The Flipper Zero device is popular in the hacker community, which can connect to radios like Bluetooth, RFID, NFC, a sub-1GHz transceiver, and Infrared. The device could act like a key fob or an entry pass, but could also be used in cases like spamming nearby iPhones.

Meanwhile, the new Flipper One device relies on network connectivity through 2x Gigabit Ethernet, USB Ethernet (5 Gbps), and Wi-Fi 6E (2.4/5/6 GHz). What’s more, the device has an M.2 port, which can be used to connect a modem for 5G connectivity or other devices like SDR modules, AI accelerators, SSDs (NVMe or SATA), and Wi-Fi cards via adapters. The device is still in development, and the company is merely announcing the project at the moment.

The device will run two processors along with an 8GB RAM. The first is an eight-core RK3576 chip that runs open Linux along with a Mali-G52 GPU and an NPU to run local AI models.

Image Credits: Flipper Devices Image Credits:Flipper Devices

The company said that it worked with open-source software consulting firm Collabora to push this chip’s support into the mainline Linux Kernel, so anyone can download it from Kernel.org and tinker with it.

The second chip is a two-core Raspberry Pi RP2350 microcontroller. This powers display, buttons, and touchpad, LEDs, and the power subsystem, so even when the Linux part of the device is turned off, users can operate the device.

Flipper Devices CEO Pavel Zhovner said that the company is also developing its own Linux-based flavour. He said in a blog that while Raspberry Pi OS is fluid and he enjoys using it, it is hard to do a clean factory reset after installing packages unless you re-flash the SD card for a new project. The Flipper OS, which is currently in a concept stage, will allow users to access profiles with different pre-configured packages and settings. By doing this, users can play around with software and go back to a clean copy without swapping or flashing SD cards.

As part of the development, the company is also making a FlipCTL interface to control small screen LCDs on devices like Flipper One with D-pad and touch controls.

Image Credits: Flipper Devices Image Credits:Flipper Devices

The company said that with network connection drivers, users can use Flipper One as a router, a VPN gateway, or a bridge between. Plus, they can plug in a monitor, along with a keyboard and a mouse through a USB Hub to make for a Linux desktop or use it as an on-the-go media box through its HDMI 2.1 port with support for 4K streaming at 120Hz. Users can also run local AI models to operate the device, generate configs, and get useful tips without an internet connection.

Flipper Devices has just announced the device and overall roadmap of how it could be used. There are still a lot of missing software pieces to enable all the mentioned features. For instance, NPU for AI and hardware video decoding lacks mainline kernel support. Both FlipperOS and FlipperCTL are concepts at the moment. The team has yet to train offline LLMs that would help users with configs.

The company is inviting developers to join the community and develop parts of the software that could be used to ship the final device. It said that the final details of the consumer launch would be announced in the future, but the device is likely to cost less than $350 for the base configuration without cellular modules.

Rising gas prices have been a recurring headline since the Trump administration started its war against Iran in late February, but it’s not the only commodity affected by the conflict. Around 10% of the world’s aluminum is made in the Gulf region, so prices of the metal have reached levels not seen in the last several decades.

Even before the war in Iran, the U.S. government had flagged aluminum as a critical mineral. A large share of U.S. demand for aluminum is met by imports, and much of the metal the country does produce is recycled. For recycling startups, it’s a good time to be in business.

“Aluminum might be 1% of the garbage stream, but it often trades for over $1,000 per ton,” Matanya Horowitz, CTO at waste sorting startup Amp, told TechCrunch. “It actually ends up being one of the most significant individual commodities.”

Aluminum is one of the most recycled materials in the U.S., but even then, only about 20% is recovered, according to the EPA. Waste sorting startups have been pitching AI as a way to improve those figures.

Sortera, a metals recycling startup, recently opened its second facility in Tennessee, the company exclusively told TechCrunch. The new site doubles the company’s processing capacity to 240 million pounds, of which 90% to 100% is aluminum. That’s a sizable fraction of the 4.3 million metric tons the U.S. used last year.

The Indiana-based startup focuses on sorting aluminum scrap. It uses a range of different sensors, including lasers, cameras and X-ray fluorescence, to feed AI algorithms that classify each potato chip-sized piece of scrap to identify the specific grade of aluminum. By separating the grades at higher accuracy, Sortera can make more profit per pound.

Amp has taken a different approach, using an AI-powered sorting system to sift through both recycling and general waste streams.

This system uses sensors, including visible light and infrared cameras, to identify everything from wrappers to foil, and differentiate plastics from aluminum. As the waste stream flows through the system on conveyor belts, robotic arms and puffers pluck or blow the materials into different bins. Amp says its system is over 90% accurate at recovering specific materials, including aluminum.

“Half of the aluminum in a metro area — in places with successful recycling programs — are just in the garbage, not even touching the recycling system,” Horowitz said. For the metals industry, recycling facilities like the kind being built by Sortera and Amp could bolster supplies of a critical mineral used through the economy.

“These types of projects are some of the biggest sources of domestically produced aluminum that are coming online in a given year,” he said.