Former Tesla executive Drew Baglino has discreetly launched a heat pump startup, as discovered by TechCrunch.

This marks Baglino’s second company established within two years of his departure from Tesla. Sources have validated the startup’s establishment, named Sadi Thermal Machines, and TechCrunch has examined company documents from Delaware and California.

Sadi was incorporated in June 2025 and shares its headquarters in Scotts Valley, California, with Heron Power, another startup initiated by Baglino, which produces solid-state transformers, as indicated by company records.

The company’s name seemingly pays homage to Nicolas Léonard Sadi Carnot, a groundbreaking French physicist whose research laid the groundwork for contemporary thermodynamics, encompassing the internal-combustion engine and the heat pump. Little information is available regarding Sadi Thermal Machines, but it seems to employ several individuals with prior Tesla experience, per a LinkedIn review and a source acquainted with the startup.

TechCrunch was unable to make contact with Baglino or the PR agency representing Heron Power. This article will be updated upon their response.

Before launching Heron Power and Sadi Thermal Machines, Baglino dedicated nearly two decades to Tesla, contributing to projects ranging from the original Roadster to the Powerwall and Powerpack energy-storage solutions. When Baglino exited Tesla in April 2024, he had advanced to senior vice president, managing the development of the company’s core energy technologies, including electric motors, batteries, and power electronics.

Throughout his tenure, Baglino focused on Tesla’s heat pumps. He is acknowledged as an inventor on a patent pertaining to a thermal management system that utilizes two coolant loops, one for cooling the battery and another for cooling drivetrain components.

To regulate the two loops, Baglino and his colleagues designed three-way and four-way valves that provide a more sophisticated control of the temperature for various components in an EV’s thermal management system. This mechanism enables Tesla to capture heat from the traction motor and repurpose it to preheat the battery, ensuring optimal performance during rapid charging in cold conditions.

The patent outlines some of the design principles underpinning Tesla’s “octovalve” system, first introduced in the Model Y. The heat pump utilized in that vehicle regulates the temperatures of the cabin, batteries, and motors, all within a compact size roughly comparable to a suitcase.

Upon its launch, Tesla’s octovalve system was more sophisticated than those of its rivals, and for a time, the company contemplated developing a heat pump for residential and commercial applications. Tesla executives, including Baglino and CEO Elon Musk, speculated about such a system during an earnings call in 2022, discussing a heat pump capable of managing both HVAC and water heating.

“From a mission perspective, it’s very aligned,” Baglino remarked. “We have gained significant insights into creating capable and reliable heat pumps that operate under all environmental conditions and are eager to tackle that challenge in the future. Let me clarify, it’s certainly in line with our mission to expedite the transition to sustainable energy.”

He noted that creating a heat pump for homes would be less challenging. “Much harder in a vehicle,” he explained. “It’s heavily constrained concerning mass, volume, and energy.”

However, Musk then added his typical caveat. “It’s something we will pursue, but we’re not committing to a timeline at this juncture,” he stated. Tesla has yet to unveil a residential HVAC or water heating system.

“People should pursue it regardless,” Baglino mentioned during the call. With Sadi Thermal Machines, he seems ready to fulfill that assertion.

This morning, OpenAI CEO Sam Altman took the stand to respond to a lawsuit from his former co-founder Elon Musk that questions OpenAI’s corporate framework.

Immediately, Altman was questioned about Musk’s claim that OpenAI’s other founders “took a charity” when they set up a for-profit branch to commercialize products based on the company’s AI technologies.

“It’s hard to even comprehend that narrative,” Altman remarked following a brief silence. “We established one of the largest philanthropic organizations globally. This foundation is accomplishing remarkable work and will achieve much more.”

Musk’s legal team has emphasized that OpenAI’s foundation, now valued at around $200 billion, did not have full-time staff until recently. OpenAI board chair Bret Taylor testified today that this was solely due to the difficulties in converting OpenAI equity into liquid assets, a feat accomplished with the recent restructuring in 2025.

The primary issue raised by Musk’s attorneys is whether the company’s pledge to safety had been neglected as it expanded commercially. However, Altman noted that in 2017, during a key moment when the founders struggled to secure funding for their AI models, Musk’s “particular plans on safety caused me concern.”

He recounted a “notably troubling moment” in the discussion when Musk was queried about what would transpire if he passed away while managing a hypothetical for-profit OpenAI. According to Altman, Musk stated, “Perhaps OpenAI should go to my children.”

Altman expressed that Musk’s fixation on controlling the initial for-profit raised alarms because OpenAI was committed to preventing advanced AI from being monopolized by an individual, and Altman, with his background in managing the well-known startup accelerator Y Combinator, was aware that “founders with control typically did not relinquish it.”

Altman also testified that Musk’s management style, which may have been effective for engineering and production, did not translate well to OpenAI.

“I don’t think Mr. Musk grasped how to effectively run a strong research institution,” Altman stated. “He demotivated some of our most crucial researchers. He had required Greg and Ilya to create a list of the researchers, outline their achievements, and rank them, resulting in significant damage to the organization’s culture for an extended period.”

In fact, Altman positioned himself as a defender of the “sweat equity” of fellow co-founders Greg Brockman and Ilya Sutskever, the individuals effectively managing OpenAI at the time while Musk and Altman were engaged in other endeavors.

Following that unresolved conflict, Musk eventually departed from OpenAI’s board and initiated competing AI initiatives at Tesla and his own AI venture, xAI. Nonetheless, Altman remained in contact with the unpredictable entrepreneur, providing updates on OpenAI’s progress and seeking his financial support and counsel.

OpenAI’s attorneys pointed out that Musk had been kept informed and invited to partake in the investments that his lawsuits now allege corrupted the non-profit.

During a discussion regarding a Microsoft investment in OpenAI in 2018, Altman shared that “contrary to many meetings with Mr. Musk, this was a positive meeting,” during which Musk engaged in a “lengthy conversation sharing memes from his phone.”

Amidst a rush from investors to acquire shares in various AI firms, Anthropic recently revised its website to alert investors that numerous private and secondary investment platforms claiming to provide access to the AI company’s shares are not authorized to do so.

The firm listed Open Doors Partners, Unicorns Exchange, Pachamama Capital, Lionheart Ventures, Hiive (new offerings), Forge Global (new offerings), Sydecar, and Upmarket as entities lacking the authorization to facilitate the buying or selling of its shares.

“Any transaction or transfer of Anthropic stock or any interest in Anthropic stock presented by these companies is void and will not be recorded in our books and records,” states the company’s support page.

When contacted for a statement, Forge Global asserted that it had been mistakenly included. “We are collaborating with Anthropic to have Forge’s name removed from this announcement,” the platform informed TechCrunch. “Forge does not facilitate transactions in any private company’s shares without the explicit consent of the company.”

In contrast, Sydecar indicated that it operates solely in an administrative role. “The company does not engage in buying or selling securities or soliciting transactions in private companies. Additionally, Sydecar mandates sponsors to verify that they have reviewed all pertinent documents regarding share transferability and that they have the necessary approvals and consents from the company,” the company stated in an email.

Anthropic’s announcement coincides with an increase in the number of investment platforms providing access to shares of AI companies (and their subsequent growth) through secondary markets where current shareholders offload their shares, “tokenized” securities, special purpose vehicles (SPVs), or secondary market assets.

The company, reportedly looking to secure additional funding at a valuation of $900 billion, has been in high demand, with some secondary market brokers telling TechCrunch last month that it’s among the “toughest” stocks to acquire.

“Anthropic is correct to seriously consider the issues surrounding unauthorized share sales and investment scams,” said Hiive spokesperson Dakota Betts in an emailed statement. “We share those concerns. They are a significant reason why Hiive made substantial investments in legal, compliance, and diligence infrastructure from the outset, and every share transfer managed by Hiive is approved by the issuer.”

Over the past year, a number of cryptocurrency firms, including crypto exchange OKX, have developed investment products offering exposure to AI firms. These typically manifest as pre-IPO perpetual futures contracts, which are derivative instruments that monitor the value of private companies on secondary markets but do not grant ownership of actual shares.

SPVs differentiate from those derivative products, providing investors with the opportunity to acquire shares of an entity that holds at least a portion of Anthropic. That equity might originate from an official investor or could have been obtained when an investor is compelled to sell their holdings, as occurred during the FTX bankruptcy. In other instances, the equity claim could be entirely fictitious.

Anthropic states that both its preferred and common stock are subject to transfer limitations, implying that any sale or transfer of shares not sanctioned by its board will be regarded as invalid. Anthropic asserts that any third-party platform (specifically SPVs and retail investment firms) that claims to sell its shares directly or via forward contracts is unauthorized to do so.

“We do not allow special purpose vehicles (SPVs) to acquire Anthropic stock, and any share transfers to an SPV are considered void under our transfer restrictions,” the company’s blog indicates. “Proposals to invest in Anthropic’s past or future financing rounds through an SPV are forbidden.”

Note: This story was updated to include comments from Hiive and Sydecar.

Potholes are an annoying issue — just consult scooter company Lime, which identified them as an official threat to its operations in its IPO documentation last week.

The history is full of assertions that technology can assist in addressing or mitigating the pothole issue, yet they continue to exist. However, as vehicles become more equipped with sophisticated sensors, they are transforming into a resource that can promptly notify municipalities about potholes and other urban challenges.

Recently, Waymo and Waze revealed a pilot initiative to exchange pothole information with local authorities. Now, fleet management firm Samsara claims it is enhancing that concept with its own AI-driven solution termed “Ground Intelligence.”

Samsara has invested the past ten years providing its clients with cameras for installation inside millions of trucks for driver oversight, theft deterrence, and assistance with liability disputes. The San Francisco-based firm has utilized all that data to develop its own model capable of identifying various pothole types and assessing their rate of deterioration.

The premise is that trucks outfitted with Samsara technology are far more widespread than Waymo’s robotaxi fleet, which currently comprises just about 3,000 vehicles. Even as that figure increases, Samsara contends it will be able to gather more data and, importantly, more repeat data from the same sites that illustrate how potholes evolve over time.

Samsara believes this information will be beneficial to municipalities — the company announced on Tuesday that it has several cities under contract and that Chicago is beginning as a new client — and that it will be the initial component of a series of insights and data points to be provided in Ground Intelligence. Additional prospective features include identifying graffiti, damaged guardrails, low-hanging power lines, or essentially “anything that we can observe that is pertinent to a city, or also to the private sector,” stated Samsara’s senior vice president of product, Johan Land.

Typically, Land explained, cities must either send out personnel or sift through numerous 311 calls to locate these issues. It creates a lot of background noise. Samsara’s proposition is that it can provide the critical information swiftly due to the numerous commercial trucks and vans currently utilizing its cameras.

Ground Intelligence functions as a dashboard. It automatically fills in alerts on a map regarding emerging potholes and other potential concerns. It also enables cities to access anonymized footage from vehicle cameras to validate citizen reports about fallen street signs, blocked sewers, or other public infrastructure issues.

“That’s the brilliance here; it transforms a process that was reactive into a proactive one,” Land stated. “That implies you won’t just address one pothole. You organize it: ‘I know where all the potholes are in this vicinity. I head out and repair them individually, in one go.'”

Samsara is also exploring additional methods to utilize this evolving municipal surveillance framework it has created. On Tuesday, it introduced a product named Waste Intelligence, which simplifies the process for waste management firms to rapidly confirm whether their clients’ refuse or recycling has been collected. Samsara also unveiled a “ridership management” solution, which can assist in alerting bus drivers to “unexpected boarding occurrences,” or generating a “digital manifest” for school buses.

During Google’s virtual Android Show: I/O Edition event held on Tuesday, the tech powerhouse revealed a collection of future updates and functionalities, featuring enhanced Gemini Intelligence capabilities; fresh hardware named Googlebook; alongside various improvements to Android, such as vibe-coded widgets, Gemini for Chrome, and new emojis.

These features will be rolled out prior to Google’s yearly developer conference scheduled for this month, which will primarily concentrate on AI, similar to last year’s focus.

Below, we’ve gathered all of Google’s announcements in one location for your convenience.

Googlebook

Google presented the Googlebook, its latest range of laptops designed with Gemini at the heart of their functionality. The tech giant is collaborating with partners such as Acer, Asus, Dell, HP, and Lenovo to manufacture the inaugural Googlebooks in various models and configurations.

The company stated that the Googlebooks, which are set to launch this fall, represent the first laptops engineered from scratch for Gemini Intelligence, offering personalized and proactive assistance.

These laptops will come equipped with “Magic Pointer,” a novel cursor integrated with Gemini; compatibility with Android smartphones that enables users to access apps from their phones directly on their Googlebook; the ability to create custom widgets; and additional features.

Google introduced a new “Create My Widget” capability enabling users to vibe-code their personalized widgets. This functionality will initially launch on the latest Samsung Galaxy and Google Pixel phones this summer.

Users will have the ability to generate widgets by articulating their desires in natural language. For instance, one might request the feature to “suggest three high-protein meal prep recipes every week” to obtain a tailored dashboard that can be added and resized on the home screen.

Updates to Android Auto

Android Auto is receiving a makeover that introduces greater personalization, widgets, and an edge-to-edge interface designed to fit any display, whether it’s an ultra-wide perspective, circular, or uniquely shaped. Users can integrate widgets that enable them to view essential information at a glance, even when navigation is active.

Media applications such as YouTube Music and Spotify are also undergoing interface redesigns aimed at simplifying their usage in vehicles.

Moreover, for the first time on Android Auto, you’ll soon have the option to watch videos on platforms like YouTube in 60 fps full HD in compatible vehicles starting this year, commencing with brands like BMW, Ford, Genesis, Hyundai, Kia, Mahindra, Mercedes-Benz, Renault, Škoda, Tata, and Volvo.

Google announced that Gemini is now widely available on Android Auto, allowing drivers to ask questions, brainstorm concepts, or learn about subjects hands-free while driving. The company also mentioned that users will gain the capability to place food orders from the car, initiating with DoorDash.

3D Emoji

Google declared that all 4,000 Android emojis have been enhanced to better reflect their actual representations. Launching later this year, the revamped emojis will be crafted to seem less flat and embody more expressiveness and realism.

Features for creators

Android is introducing a “Screen Reactions” feature that captures you and your screen simultaneously, a format commonly seen on TikTok and Instagram Reels. This capability will initially be available on Pixel devices this summer.

Google has also collaborated with Meta to introduce premium Instagram features to its Android devices, including Ultra HDR, native stabilization, and night mode. Additionally, the company has optimized the process from capture to upload to ensure your photos and videos retain clarity upon posting.

The tech giant is providing new tools within Meta’s Edit app, exclusively on Android, including “smart enhance” for photo upscaling and “sound separation” for amplifying and removing audio.

Gemini’s agentic features

With Gemini Intelligence, the assistant will be capable of using data from one application to execute multistep tasks across various apps. For example, users could photograph an event flyer and request the assistant to find that event on platforms like Expedia.

This feature allows users to summon the assistant with their grocery list displayed and ask it to compile a shopping cart based on those items within the selected shopping application.

Gemini in Chrome

The company is now launching Gemini in Chrome for Android, following earlier releases on iOS and desktop, enabling users to summarize content or inquire about what they observe on the web page. Android users will also gain access to the experimental auto-browse feature, capable of navigating websites and completing actions like booking a ticket on a user’s behalf.

Form filling

Gemini will utilize data from Personal Intelligence to assist users in filling out intricate forms on mobile devices through an opt-in feature.

Dictation

Google has introduced a new functionality within Gboard called Rambler, which converts your spoken words into neat text — similar to applications like Wispr Flow and Monologue. This feature eliminates filler phrases like “ums” and “ahs” and also captures the intent behind statements, such as “Let’s meet at 3 p.m. … um, 2 p.m.” and outputs “Let’s meet at 2 p.m.” as the final text.

Extended AirDrop support

Last year, Google found a method for Pixel devices to share images and files with iPhones by enabling Quick Share to integrate with AirDrop. The company announced that this year, the feature will be accessible to users from other smartphone brands, including Samsung, Oppo, OnePlus, Vivo, Xiaomi, and Honor.

Furthermore, if you lack a compatible device, you can utilize Quick Share on any device to generate a QR code, which will allow you to share files with an iPhone via the cloud.

Additionally, Google stated that users will soon have the ability to access Quick Share within apps such as WhatsApp.

Easier iPhone-to-Android transfer

Google is set to roll out a new transfer option from iOS to Android, enabling you to bring over your passwords, images, messages, favorite apps, contacts, eSIM, and your home screen arrangement from your iPhone to your new Android device. This functionality is anticipated to launch on Samsung Galaxy and Google Pixel devices this year.

Pause Point

A new Android feature will permit you to pause when launching an app that you’ve flagged as one of your distractions. Before you can browse, game, or engage in any time-consuming activities, you’ll need to take a 10-second interval to consider alternative choices you could select instead — such as launching your Google Play Books app for reading. Additionally, you can optionally set a timer that will cut you off from using the app before you dive in.

Improved threat-protection features

Google is also rolling out its default theft protections to all Android users globally following initial trials in Brazil. These protections will automatically activate on all new Android 17 devices, as well as devices that have been freshly reset or updated to the latest operating system.

When enabled, features such as Remote Lock and Theft Detection Lock will be turned on by default, and Google is limiting the number of attempts a thief can make to guess a user’s PIN or password. Additionally, there will be extended intervals between failed attempts. Law enforcement officials will now have the ability to access the device’s IMEI from the lock screen on Android 12 and above, thus allowing them to verify device ownership more efficiently if stolen.

Moreover, theft protections will be available for devices running Android 10 and upward in select regions, including Argentina, Chile, Colombia, Mexico, and the U.K.

Pixel users with up-to-date software and Advanced Protection Mode enabled will now benefit from Intrusion Logging, a security feature designed to assist in investigating suspected spyware attacks and device compromises.

Google is introducing a new optional feature in Android aimed at assisting security researchers in examining spyware incidents.

This feature, termed “Intrusion Logging,” is part of Android’s Advanced Protection Mode, introduced last year. This opt-in security mode activates specific functionalities to enhance device security against hacking attempts. Advanced Protection Mode seeks to mitigate government spyware attacks and forensic tools used by law enforcement to extract information from individuals’ phones.

These two forms of attacks can occur in tandem. In at least one reported instance in Serbia, authorities utilized a forensic tool from Cellebrite to access a device, subsequently installing spyware to maintain surveillance on the target. 

The release of Intrusion Logging marks the first occasion a device manufacturer has developed a feature specifically intended to aid security researchers in probing spyware incidents. To make this happen, Android’s Intrusion Logging generates a novel log type that catalogs errors and gathers evidence when irregularities occur within the software, offering insights into suspected spyware attacks. 

Amnesty International, which collaborated with Google to create this feature, described Intrusion Logging as “a significant transformation in the quantity and quality of forensic information accessible on Android devices.”

“Historically, forensic investigations relied on logs that were never intended for intrusion detection,” Amnesty noted in a blog entry providing detailed information on how Intrusion Logging operates. Consequently, previous logs lacked utility for researchers as they were stored temporarily and frequently overwritten, effectively obliterating any potential evidence of attacks.

Donncha Ó Cearbhaill, head of Amnesty’s Security Lab, indicated to TechCrunch that Android’s technical constraints “have hindered in-depth analysis of system logs and files for indications of compromise, unlike iOS.”

“These constraints have prevented us from reliably recognizing known threats against Android,” stated Ó Cearbhaill, who has investigated numerous spyware abuse cases globally over the years. 

The capability to more effectively recognize spyware attacks is expected to enhance with Intrusion Logging. Google announced this feature a year ago, but is now initiating its deployment. In a blog post on Tuesday, Google stated that Intrusion Logging “is currently being rolled out to all devices operating on the Android 16 December update and newer.”

How Intrusion Logging operates

Intrusion Logging monitors events related to security and possible breaches. Initially, the feature compiles and records logs daily, storing them securely in a users’ Google account in the cloud. By uploading logs to the cloud, it potentially shields evidence of device compromise from being erased by spyware. The logs are also encrypted to ensure that only the user can access them and share them with investigators, preventing Google from accessing the data.

The events tracked by Intrusion Logging include instances of phone unlocking; installation and uninstallation of applications; connections made to websites and servers; whether someone connected via Android Debug Bridge, a tool that enables a computer or forensic device like Cellebrite to link to an Android device; and efforts to erase logs associated with these events, which may suggest attempts to conceal evidence of an attack. 

In the context of a spyware breach, these logs can assist investigators in understanding when and how authorities might have compromised or forcibly accessed an individual’s device, linked it to a forensic tool, or used it to implant spyware or stalkerware. The logs may also reveal if a device connected to a malicious site aimed at exploiting visiting devices or accessed servers intended for data extraction from the phone. 

Although it’s a positive development, Intrusion Logging has certain limitations. Currently, in addition to needing to activate Advanced Protection Mode, this feature requires the latest Android software version, is exclusively available for Google-produced Pixel devices, and necessitates linking the device to a Google account. Intrusion Logging maintains records of browsing history and connections, which some individuals may hesitate to share with investigators. 

Google asserts that Advanced Protection Mode and Intrusion Logging cater to individuals who might be vulnerable to spyware and forensic device attacks, including human rights advocates, activists, journalists, and dissidents. Advanced Protection Mode is akin to Lockdown Mode for Apple devices, also designed for high-risk users and regarded as an effective strategy against spyware. 

As recently as March, Apple stated it has never found a successful attack against users who enabled Lockdown Mode. In 2023, security researchers from Citizen Lab reported that Lockdown Mode successfully thwarted an attempt to compromise a target with NSO’s spyware. 

In its blog post, Amnesty provided comprehensive instructions on retrieving the logs if a user suspects or has been alerted to being targeted by spyware. Apple, Google, and Meta have been notifying users about threats for years, which researchers indicate has been vital for identifying and exposing abuse cases.

On Tuesday, Anthropic revealed that it is introducing a range of new chatbot functionalities aimed at delivering automated support to legal firms. These enhancements broaden Claude for Legal — the law-focused plug-in that was launched earlier this year — providing users with an updated suite of legal plug-ins and MCP connectors tailored for particular legal domains.

These latest tools emerge amidst fierce competition in the legal AI arena. In March, the AI law startup Harvey, which leverages agentic AI for legal workflow automation, secured $200 million at a valuation of $11 billion. Just last month, another startup, Legora, completed a $600 million Series D funding round and kicked off an eye-catching advertising campaign featuring Jude Law. Legora provides services akin to those of Harvey — automated solutions designed to streamline the often convoluted legal processes that previously required entire teams of people.

The tools from Anthropic aim to assist legal firms in automating specific administrative tasks — such as document searching and reviewing, accessing case law, preparing for depositions, drafting documents, and other related areas. The plug-ins — which encompass a collection of functions and automated tools — are crafted to function across various legal domains like commercial, privacy, corporate, employment, product, and AI governance, as stated by Anthropic.

Additionally, Anthropic is providing several model context protocol connectors. MCPs link distinct data sources and external systems to AI models, enabling direct interaction with them. In this context, the new MCP connectors integrate Claude into numerous software applications routinely utilized by law firms — such as document management tools like Docusign and file search systems like Box. Legal research platforms like Thomson Reuters (which runs Westlaw) can also be integrated.

According to the company, the new connectors and plug-ins will be accessible to all Claude subscribers. These new functionalities also build upon previous plug-ins tailored for the legal sector that the company introduced in February.

“The legal industry is under increasing pressure to embrace AI, and the firms and in-house teams that take initiative are advancing quickly,” stated a company spokesperson. “Claude is making a significant move into knowledge work, with the legal field emerging as one of its most important and rapidly growing sectors.”

As AI companies have attempted to attract law firms, failures related to AI have led to notable complications in court. Numerous lawyers have been discovered using AI to create documents filled with errors, including at least one prominent law firm. Last year, California imposed a pioneering fine on an attorney who used ChatGPT to draft an appeal filled with fabricated quotes. Federal judges have also been detected using it to prepare rulings, a trend that attracted the attention of congressional leaders last year. Furthermore, it is reported that AI-generated lawsuits are congesting the judicial system — inundating courts with piles of poorly argued legal “slop.”

On Tuesday, Google introduced a new “Create My Widget” feature for Android, enabling users to code their personalized widgets. This feature will debut on the latest Samsung Galaxy and Google Pixel devices this summer.

Users can create a widget by articulating their desires in natural language. For instance, one might request the feature to “recommend three high-protein meal prep recipes every week” to generate a custom dashboard that can be added and resized on the home screen.

Alternatively, if you’re a cyclist focused solely on wind speed and rain, you can create a weather widget that displays only those specific metrics on your home screen.

Gemini also has the ability to retrieve information from the internet and integrate with Google applications such as Gmail and Calendar to assemble a unified, personalized dashboard. For example, if you’re organizing a family reunion in Berlin, it can compile your flight and hotel information, highlight restaurant bookings, and even include a countdown.

This feature represents Google’s latest effort to integrate generative AI more deeply into the Android ecosystem, as technology firms compete to enhance customization tools for everyday users.

“It’s akin to asking your personal assistant a question and having them repeatedly deliver the answer,” stated Ben Greenwood, director, PM, Android Core Experiences, during a briefing with reporters. “Consider it as querying Gemini about the world, its understanding of current events, and aspects related to your personal data. These two categories unlock a wealth of use cases that we’re incredibly enthusiastic about.”

The company announced this new functionality alongside the introduction of Gemini Intelligence, which will offer more utilities like advanced autofill, a voice dictation feature powered by AI for Gboard, and other capabilities.

Google unveiled several new AI functionalities under the Gemini Intelligence brand during its Android Show: I/O Edition event on Tuesday. Among these enhancements are the capabilities for AI to perform tasks across different applications, navigate the web, fill in forms, articulate speech, and even enable you to vibe-code customized Android widgets.

Gemini becomes more robust

Earlier this year, at the Samsung Galaxy S26 launch, the company introduced some agentic abilities like ordering food or securing a ride for Gemini. During that event, Google revealed that Gemini would soon gain the ability to carry out more intricate tasks, such as reserving a front-row bike for a spin class, retrieving a class syllabus from Gmail, and searching for books relevant to that subject.

Now, Google’s AI assistant will handle a multi-step process, such as copying a grocery list from your notes app and adding items to your shopping app’s cart. To access this function, you’ll press the power button on your phone and state the task. Meanwhile, the content displayed on the phone’s screen provides context for the assistant. Google emphasized that Gemini will wait for your final approval before completing the checkout.

Moreover, a feature introduced in January allowed Gemini to browse the web and complete tasks such as scheduling appointments, as part of an experimental rollout. Today, Google announced that this auto-browse functionality will also be available on Android.

By late June, Android devices will see the addition of Gemini in Chrome, an AI feature designed to assist users in summarizing content or posing questions about web page materials, similar to the existing functionality of Gemini in Chrome on desktops.

Additionally, Gemini will be able to fill out forms on your behalf by acquiring personal information through Personal Intelligence. (Google mentioned that this feature is opt-in and can be disabled through settings at any time.)

Furthermore, Gemini will integrate with Android’s Gboard keyboard. Google is leveraging Gemini’s multimodal capabilities by introducing a feature called Rambler within Gboard, akin to those found in other AI-driven dictation applications. This feature will allow users to express themselves naturally, transcribe their speech, and refine it by eliminating filler phrases.

Vibe-coding applications are gaining momentum, and Google aims to offer Android users an experience of this innovation as well.

The company is implementing a method for users to create Android widgets by articulating them using natural language. For instance, users can generate a meal-planning widget with query phrases like, “Recommend three high-protein meal prep recipes weekly.”

The concept of widget creation is not new to Gemini. Notably, the hardware start-up Nothing also launched a comparable tool the previous year.

Google indicated that Gemini Intelligence will adhere to the company’s Material 3 expressive design standards in its functionalities.

The company stated that these AI-driven features will initially launch on the latest Samsung Galaxy and Google Pixel devices this summer and will subsequently be accessible on other Android devices later this year.