Your smartphone has a pile of apps. OpenAI wants to replace all of them with one AI agent that just gets things done. That’s the vision behind the company’s plans to build its own smartphone, complete with a custom processor co-developed with MediaTek and Qualcomm, as first reported by analyst Ming-Chi Kuo on X.

And Sam Altman seems to agree. In a post on X, the OpenAI CEO wrote, “feels like a good time to seriously rethink how operating systems and user interfaces are designed.” That is not a subtle hint.

Why would OpenAI want to make a phone?

We have seen earlier attempts at developing truly agentic AI in the form of Rabbit, Humane AI Pin, and other AI devices. However, those devices lacked the tight integration with our phones, apps, and services, resulting in failure. It seems that OpenAI wants to sidestep the limitation by creating its own phone to provide users with a true AI assistant. 

There are three solid reasons. First, to deliver a truly comprehensive AI agent experience, OpenAI needs full control over both the software and the hardware. Relying on Android or iOS means playing by someone else’s rules.

Second, your smartphone knows more about you than any other device. It tracks your location, your habits, and your daily context in real time. That kind of data is gold for an AI agent trying to anticipate your needs before you even ask.

Third, smartphones are and will remain the biggest device category on the planet. If OpenAI wants to scale, this is where it needs to be. 

Advertisement

How will the AI actually work on this phone?

According to Ming-Chi Kuo, the new OpenAI smartphone will work on a two-layer system. The phone will handle lighter tasks on-device, like understanding your context, managing memory, and running smaller AI models. Heavier tasks get offloaded to the cloud. 

It’s similar to what Apple does with its iPhone and Private Cloud Compute, but OpenAI has the benefit of an actually working artificial intelligence model and not the disaster Apple calls Apple Intelligence.

On the business side, OpenAI is likely looking at bundling hardware with subscriptions, similar to how Apple bundles services, while also building a developer ecosystem around its AI agents.

Who is helping OpenAI build this thing?

Mr. Kuo reports that MediaTek and Qualcomm are the processor co-development partners, while Luxshare is the exclusive system co-design and manufacturing partner. Luxshare is particularly interesting here. 

According to Kuo, the company has long tried to challenge Hon Hai’s (read Foxconn) dominant position in Apple’s supply chain without much success. This project gives Luxshare an early foothold in what could be the next major smartphone generation, and that is a big deal for the company.

Advertisement

2028 feels far away, but if OpenAI pulls this off, the smartphone you are using today might look very different in the near future.

“Lights, camera, action!” might have been the call when recording back in the day, but for an awesome three-dimensional viewing experience, you might try yelling “Mist, Mirrors, Laser!” and following in the footsteps of [Ancient]’s latest adventure in voxel displays, which is also embedded below.

He starts with a naive demonstration: take a laser projector and toss an image into a flat cloud of mist. That demonstrates that yes, the mist does resolve an image, and that the viewing angle is very poor– that is, brightness drops off sharply when you’re out of line from the projector. In this case, that’s a good thing! It means more angles can be projected into that mist for a three-dimensional, hologram effect.

The optical train gets folded up, probably to make this fit on a tabletop: first, an array of flat mirrors in front of the projector splits the image from the projector into multiple viewpoints, which are each bounced to a second flat mirror that sends the image into the fog bank.

Some might call the resulting image a hologram; others might complain that that’s technically something totally different, and that this volumetric display is just all smoke and mirrors. We can hope that [Ancient] sees fit to share more details, like the software stack needed to generate the video feed– though it’s likely using a version of the same software as his last volumetric display, which used the same laser but whose point cloud was made from a bubblegram rather than an actual cloud. With a lot more points, though, the resolution is amazing in comparison, at the cost of appearing fuzzy at the edges. Unfortunately, we do not see the display in this demo run DOOM, as one of his previous projects did.

This video is more of a demo than a how-to, but it’s a heck of an impressive demo. If you don’t feel like watching the assembly, jump right to 9:00 to be impressed. It comes across a lot better on video than in the screenshot.

Amazon’s podcasting business has transformed over the past six months, according to The New York Times.

Back in August 2025, the company reportedly eliminated more than 100 jobs from its podcast studio Wondery. At the time, Amazon insisted it was not shutting Wondery down, and that appears to be technically true — it still uses the Wondery brand.

But the NYT said Amazon “took a sledgehammer” to the studio. Audio-only podcasts now operate under Audible, while a new department called Creator Services works with on-camera celebrities like Dax Sheperd, Keke Palmer, and Jason and Travis Kelce.

For example, the company said it’s creating an “expanding universe” around the Kelce brothers’ “New Heights,” with monetization plans that go far beyond standard podcast ads. There’s a new section on Amazon called Kelce Clubhouse, where fans can buy “New Heights” merchandise, watch the documentary “Kelce,” and purchase recommended products for a football-watching party.

In the words of Creator Services general manager Matt Sandler, Amazon is trying to “infuse both the content and the commerce together.”

Of course, other online creators are also betting on commerce. But according to the NYT, Amazon is the only one that “dismembered a company” to get here.

China’s most powerful governing bodies, the Chinese Communist Party Central Committee and the State Council, issued comprehensive labour rules for gig workers on Sunday, the first time the party’s highest authority has formalised protections for the more than 200 million people who deliver food, drive cars, and livestream products through online platforms. The mandate requires platforms to pay at least the local minimum wage, enforces maximum working hours after which the app must stop sending orders, mandates algorithm transparency when platform policies affect pay or task assignment, and sets a target of 2027 for broadly standardising labour practices across the platform economy. Previous regulatory efforts came from individual ministries and carried the weight of guidelines. This comes from the top, and it covers everyone: Meituan, Didi Chuxing, Alibaba’s Ele.me, JD.com, SF Express, and ten other major platform and logistics operators summoned by the Ministry of Human Resources and Social Security in February for what the government called “employment administrative guidance.”

The rules

The regulations establish several concrete protections that did not previously exist in binding form. Platforms must ensure gig workers receive at least the local minimum wage, with reasonable additional compensation for work during public holidays. Enterprises must negotiate with labour unions or worker representatives to determine maximum consecutive order-taking time and maximum daily working hours. When workers reach those limits, the system must stop dispatching new orders and send push notifications through the app reminding workers to rest. Businesses must enter into employment contracts with workers when conditions for an employment relationship are met, and for workers who fall below that threshold, they must sign written agreements specifying terms. Platform enterprises must seek worker input when formulating or revising labour rules, which must be publicly displayed for at least seven days before taking effect.

The algorithm provisions are the most significant departure from previous Chinese labour regulation and from anything the European Union or the United States has enacted for gig workers. Platforms must develop and regularly revise the algorithms that control onboarding, task assignment, piece rates, commission structures, compensation, work hours, and incentive or penalty systems. They must consider labour union or worker representative opinions when designing those algorithms and must agree to negotiations if unions request them. They must furnish the information and materials necessary for those negotiations. This is not a transparency requirement in the Western sense, where a company publishes a report about its algorithm. It is a requirement that the algorithm itself become a subject of collective bargaining, that the code governing how a delivery rider earns a living be open to negotiation with worker representatives. Many jobs are being reshaped at the task level by AI rather than disappearing wholesale, and nowhere is that reshaping more literal than in the gig economy, where the algorithm is the manager, the dispatcher, and the payroll department.

The system

The conditions the regulations address have been documented for years. In September 2020, Renwu magazine published “Delivery Workers, Trapped in the System,” an investigation based on six months of research that became the most viral article on the Chinese internet that year. It documented how Meituan and Ele.me’s algorithms progressively shortened delivery times, forcing riders to run red lights, drive against traffic, and sprint up staircases. Per-order pay was determined by a system that factored in average daily orders, punctuality, customer ratings, and complaints, a calculation opaque to the riders whose income depended on it. In Shanghai, in the first half of 2017, one delivery rider was injured or killed every 2.5 days. In Chengdu, in the first seven months of 2018, there were roughly 10,000 traffic violations by delivery riders, 196 accidents, and 155 injuries or deaths, approximately one per day. In September 2024, a delivery rider in Hangzhou collapsed and died after working 18-hour days. A 2023 survey found that roughly half of food delivery riders earn between 4,000 and 5,999 yuan per month, $563 to $845, and only 7% earn more than 8,000 yuan.

Ele.me’s response to the Renwu investigation was to introduce a button allowing customers to “wait five extra minutes,” a gesture that was widely criticised for shifting responsibility from the platform to the consumer. Workplace surveillance under a different name is not unique to Chinese platforms. Meta has installed tracking software on American employees’ computers to monitor keystrokes and mouse movements. But the scale of algorithmic control in China’s gig economy is different. Meituan alone had 4.72 million active riders as of 2020. Didi created 30.66 million flexible job opportunities. Meituan and Ele.me together control approximately 98% of China’s food delivery market. When two companies’ algorithms govern the working conditions of six million delivery drivers, regulating those algorithms is not a labour policy niche. It is macroeconomic governance.

The context

The timing of the regulations is inseparable from China’s economic circumstances. Youth unemployment stood at 16.5% in December 2025, and some economists estimate the real figure exceeds 40% when discouraged workers and those in involuntary part-time work are included. More than 12 million university graduates are expected to enter the job market in 2026, and many of them will find their first employment on a platform. The 15th Five-Year Plan, covering 2026 to 2030, elevates consumption to a dedicated chapter for the first time, reflecting Beijing’s strategic pivot from export-led and investment-led growth toward domestic consumption. That pivot requires household income to grow, which requires the 200 million workers in flexible employment, roughly 27% of the total workforce and 43% of the urban labour force, to earn enough to spend. Gig workers who earn $563 a month and have no social insurance are not consumers who drive a consumption economy. They are a fiscal liability and a source of social instability. The regulations are labour policy, but they are also demand-side economics.

China has nearly closed the AI performance gap with the United States, spending 23 times less on AI investment to do so, and the platform companies that employ these gig workers are at the centre of that achievement. Meituan, Didi, and Alibaba are not just delivery and ride-hailing firms. They are AI companies whose logistics algorithms, recommendation engines, and autonomous delivery experiments represent some of the most advanced commercial AI deployments in the world. Beijing’s calculation is that it can impose labour costs on these platforms without destroying the innovation ecosystem, because the platforms are profitable enough to absorb them and because the alternative, 200 million workers with no protections and no purchasing power, is worse for the economy than higher delivery costs.

The comparison

The EU Platform Workers Directive, adopted in December 2024 with a transposition deadline of December 2026, takes a different approach. It establishes a rebuttable legal presumption of employment: if a platform exercises direction and control over a worker, the worker is presumed to be an employee unless the platform proves otherwise. The burden of proof is on the company. Workers cannot be fired solely by algorithm. The UK Supreme Court ruled in 2021 that Uber drivers are workers, not independent contractors, and that waiting time counts toward minimum wage calculations. California’s Assembly Bill 5 presumed gig workers were employees, but the platforms spent over $200 million campaigning for Proposition 22, which exempted them and was upheld by a state appeals court in 2023. The EU AI Act targets safety, transparency, and ethics but falls short on socio-economic impact, and the Platform Workers Directive was partly an attempt to fill that gap.

China’s approach is less categorical than the EU’s presumption of employment and more interventionist than America’s deference to corporate self-governance. It creates a spectrum: formal employment relationships requiring full contracts and benefits at one end, a middle category with written agreements and partial protections in the middle, and a regulatory floor of minimum wage, maximum hours, and algorithm transparency for everyone. The middle category is where the ambiguity lies. It gives platforms a classification that carries fewer obligations than full employment but more than the independent contractor status that Uber and DoorDash insist on in the United States. Whether that middle category becomes a genuine improvement or a loophole that platforms exploit to avoid full employment obligations depends on enforcement, and enforcement is the historical weakness of Chinese gig worker regulation. China’s AI governance framework requires all generative AI models to pass a security evaluation, and the Cyberspace Administration of China enforces those requirements rigorously. Whether the labour agencies tasked with enforcing the new gig worker rules will demonstrate the same rigour is an open question.

The test

The companies have already begun responding. SF Express set aside 200 million yuan, approximately $29 million, to increase delivery worker income. Didi announced 1.1 billion yuan in driver subsidies. Alibaba pledged to cover at least 50% of social security for delivery riders. JD.com committed to full social benefits for all full-time riders. Meituan and Ele.me both pledged enhanced social security coverage. These are not trivial commitments, but they are pledges made under regulatory pressure, and the gap between a pledge and a payslip is where previous Chinese gig worker regulations have failed. The 2021 guiding opinions from eight ministries covered labour income, safety, social security, and conflict resolution. They were widely acknowledged and narrowly implemented. Delivery times kept shrinking. Riders kept dying. Algorithms kept optimising for speed over safety.

The difference this time is the source. Ministry-level guidelines can be deprioritised. A mandate from the CPC Central Committee and the State Council cannot. The 2027 compliance deadline gives platforms 18 months to restructure their labour relationships, algorithm governance, and social insurance contributions. The enforcement mechanisms remain underspecified in the publicly available text, and that is a legitimate concern. But the political signal is unambiguous. Xi Jinping’s government has decided that the platform economy’s treatment of its workers is a problem significant enough to warrant the highest level of party intervention. Whether the intervention produces the kind of structural change that 200 million workers need, or whether it produces another round of corporate pledges that dissolve into algorithmic business-as-usual, is the test that 2027 will answer. China has written the rules. The question, as always in Chinese regulation, is who enforces them and what happens when they are broken.