The latest leap in artificial intelligence is forcing businesses to reconsider a question that goes far beyond which chatbot they should use: How much of the work itself still needs to be performed by people?

OpenAI’s newly released GPT-6 Astra is being positioned as a major advance in AI capabilities, particularly in computer use, coding and completing complex multi-step tasks. NVIDIA CEO Jensen Huang has even declared that artificial general intelligence, or AGI, has arrived with Astra — although that claim remains controversial and there is no universally accepted definition of AGI.

For businesses, however, the AGI label may be less important than what these systems can actually do.

The biggest change is the growing ability of AI to complete work rather than simply generate information. Instead of asking an AI to write an email, summarize a report or produce an idea, companies can increasingly give AI a larger objective and allow it to work through multiple steps toward completion.

That changes the economics of automation.

A company could eventually have AI handling portions of customer service, research, administrative operations, sales follow-up, software development and internal analysis with considerably less human intervention. The human role shifts from performing every step to setting objectives, reviewing results and handling the situations AI cannot reliably resolve.

That does not mean businesses should immediately replace employees with AI. It means companies should start examining their workflows differently.

The companies that gain the most from increasingly capable AI may not be the ones that simply purchase the newest model. They will be the ones that redesign their operations around what AI can now accomplish.

This is also why the arrival of more autonomous AI creates a new management challenge. OpenAI’s chief scientist has warned that increasingly capable agents could create consequences that organizations and society are not yet prepared to manage.

For executives, the message is straightforward: AI is moving from a productivity tool toward a potential digital workforce.

Businesses should be asking now which tasks can be automated, where humans must remain in control, and how employees can move toward higher-value responsibilities.

The competitive advantage may no longer come from simply having AI.

It may come from knowing how to reorganize the business around it.

NVIDIA is making a massive bet that the future of artificial intelligence will not be controlled solely by a handful of companies selling access to closed AI models.

The chip giant has agreed to acquire Hugging Face for $12.93 billion, one of NVIDIA’s largest acquisitions. Hugging Face has become a central platform for developers building, sharing and deploying open-source and open-weight AI models. More than 18 million developers, researchers and creators use the platform, while more than 200,000 companies rely on it for AI development.

For businesses, the deal matters because it points toward a future in which companies have far more choices about how they build AI.

Rather than depending entirely on expensive proprietary models, businesses can increasingly customize open models for specific tasks, run them across different cloud providers and potentially deploy them using their own infrastructure. NVIDIA says Hugging Face will remain open and will continue supporting different models, clouds and computing platforms rather than requiring NVIDIA hardware.

That could eventually make enterprise AI more flexible and less expensive.

But there is another message behind the acquisition: AI infrastructure is becoming the real battleground.

NVIDIA already dominates the chips powering modern AI. By moving deeper into the software and developer ecosystem, the company is positioning itself across more of the AI stack—from the computing hardware to the models and tools businesses use to build applications.

For business owners, this means the AI decision is becoming less about asking, “Which chatbot should we use?” and more about asking, “What AI infrastructure gives our company the greatest control, flexibility and return on investment?”

The companies that begin experimenting with customized models, AI agents and internal AI systems now may have an advantage as these technologies become cheaper and more capable.

The NVIDIA-Hugging Face deal is therefore more than a $13 billion acquisition. It is a signal that the next phase of business AI may be defined by open models, customized systems and control over the underlying AI infrastructure.

And for businesses, that could ultimately mean more powerful AI without being locked into a single vendor.

Nvidia is making one of its biggest moves beyond chips, agreeing to acquire AI platform Hugging Face for roughly $13 billion. The deal is significant because Hugging Face has become a major home for open-source AI models, datasets and applications, with more than 18 million developers and 200,000 companies using the platform. Nvidia says Hugging Face will remain open and support multiple cloud and computing platforms.

What It Means for Businesses

The acquisition points to an important shift in the AI market: businesses are increasingly looking beyond simply subscribing to a chatbot.

Open-source AI gives companies more opportunities to customize models, run AI within their own infrastructure and reduce dependence on a single AI provider. Nvidia’s investment could accelerate that trend by combining its computing infrastructure with one of the world’s largest open AI communities.

For smaller businesses, the bigger takeaway is that AI is becoming infrastructure rather than an experimental tool. Companies that build AI into sales, customer service, marketing, operations and internal workflows are likely to have more choices about which models power those systems.

But there is also a warning. Hugging Face has recently faced AI-related security concerns, while businesses are giving autonomous AI agents increasing access to company systems. Security researchers and lawmakers are now pushing for stronger controls around what AI agents can access and execute.

The business opportunity is no longer simply “use AI.” It is building an AI stack that is flexible, secure and capable of changing as better models arrive.

For business owners, that means the companies that avoid locking themselves into one AI model today may have a significant advantage tomorrow.

There is a question New York seems increasingly afraid to ask because asking it sounds insufficiently compassionate: When did someone else’s mental illness become everybody else’s responsibility to endure?

I am in in Times Square most every day. I take New York City buses. I walk these streets and like millions of New Yorkers, I have found myself trapped in public spaces with people screaming, threatening, striking out at invisible enemies, terrorizing passengers or behaving so erratically that everyone around them begins calculating where the nearest exit is. We have normalized it. Don’t engage. Move away. Change subway cars. Get off the bus. Don’t make eye contact. Give the person space. Why is the burden perpetually placed upon the person who is simply trying to get home or on about their business?

On August 31, Erin Piacenti was doing exactly that. She was 32 years old, a lawyer and vice president at Bank of America, and had just returned to work after five months of maternity leave. She was walking through Times Square after work when Pamela Cisneros, 49, stabbed her in the abdomen in what police described as a random and unprovoked attack. Cisneros had also stabbed 68-year-old Tak Kam, who survived. Piacenti did not. She left behind a husband and a five-month-old daughter.

Almost immediately, much of the story has became about the mental illness of the woman who killed her. Cisneros had suffered from serious mental-health problems for decades, according to her father. In 2019, she reportedly attempted to jump in front of a subway train. Afterward, her father began taking her to monthly hospital appointments where she received injections that he said kept her stable. In April, according to his account, she told him she could handle those appointments herself and stopped allowing him to accompany her. It remains unclear whether she continued receiving treatment.

That history explains something. It does not resurrect Erin Piacenti. That distinction has become dangerously blurred. Mental illness may explain why someone behaves irrationally. It may explain why someone cannot regulate behavior or distinguish reality from delusion. But an explanation does not make the consequences disappear for everybody else.

Erin Piacenti’s daughter will not grow up with her mother. Her husband does not somehow lose less of his wife because the attack was committed by someone with a history of mental illness. Tak Kam was not stabbed less deeply because his attacker was psychiatrically troubled. Their suffering counts too.

Yet increasingly, when something like this happens, our cultural reflex is to move almost immediately toward the suffering of the person who caused the catastrophe. What happened to her? Why wasn’t she receiving treatment? What trauma had she experienced? What diagnosis did she have? Those may be legitimate questions, but there is another question that deserves considerably more attention: Who was responsible for making sure a person known to be dangerously unstable did not reach the point where complete strangers became responsible for surviving her?

That responsibility cannot belong to a woman walking home from work. It cannot belong to a tourist coming out of the subway or passengers trapped on a bus. It cannot perpetually belong to police officers who encounter someone at the very end of a psychiatric collapse and are expected to solve in seconds what families, doctors, hospitals and institutions have failed to solve over years.

Cisneros ultimately approached police carrying two knives, refused repeated commands to drop them and continued toward officers even after Tasers were deployed. Officers shot and killed her. By the time the episode ended, an innocent woman was dead, another innocent person was wounded, Cisneros herself was dead and police officers had been placed in a confrontation that never should have reached that point.

There is also something we have become extraordinarily reluctant to discuss: families have responsibilities. Yes, an adult has rights. Yes, families cannot simply control another adult. Yes, serious psychiatric illness can make treatment extremely difficult, and relatives themselves can become exhausted, frightened and overwhelmed. Family cannot simply disappear from the discussion when someone has a known history of profound instability.

In this case, Cisneros’ father appears to have tried. He accompanied his daughter to treatment and helped make sure she received medication. When she said she could handle those appointments herself, he believed she had turned a corner. That makes this particular case more complicated, but it also exposes the enormous hole in the system. What is a family supposed to do when an adult relative with a history of serious psychiatric instability refuses help? Who follows up when someone receiving treatment that has apparently kept her stable suddenly stops appearing? At what point does somebody have the authority—and the responsibility—to say this person is not capable of simply being left alone?

We have created a remarkable vacuum of responsibility. The individual is mentally ill, therefore her responsibility may be diminished. The family cannot force treatment. Doctors face legal restrictions. Police frequently cannot intervene until particular thresholds are crossed. Everyone else is instructed to avoid confrontation and protect themselves. Responsibility evaporates precisely when it becomes most necessary, and somewhere inside that maze stands an ordinary person who never volunteered to participate in any of it. Sometimes that person ends up dead.

Times Square magnifies the problem because everybody comes through here. Workers, tourists, families, performers, hustlers, homeless people, people struggling with addiction and people experiencing profound psychiatric distress are compressed into a few blocks of public space. Those of us who actually live and work here know the choreography. You notice the person screaming before you reach the corner. You watch someone’s hands. You move farther down the platform. You decide whether to board the subway car. You quietly change seats on the bus. That is not prejudice. It is self-preservation.

We do not owe another human being the surrender of our own safety because that person is ill. We can acknowledge mental illness without making innocent strangers responsible for enduring its consequences.

Erin Piacenti had spent five months at home with her baby. Four days before her death, she wrote publicly about how grateful she was for that time. On her first day back at work, she walked through Times Square and never made it home. Her daughter is now five months old and motherless.

Erin Piacenti did not cause any of this. Neither did Tak Kam. Neither do the millions of New Yorkers who board buses, enter subway cars and walk through Times Square every day.

Mental illness needs treatment there is no denying it, but the public is entitled to something as well, and we have become strangely apologetic about saying it. We have the right to go home alive.

Tesla is showcasing its two-seat Cybercab in Austin as it pushes deeper into autonomous ride-hailing, and the vehicle’s most striking feature may be what it does not have: a traditional steering wheel. That design decision matters because it signals that Tesla is not simply introducing another car. It is trying to build a vehicle meant to function primarily as part of a transportation network rather than as a product someone buys, parks in a driveway and drives personally. If that strategy works, the economics of the car business could start to look very different.

For most of automotive history, the business model has been simple. A car company designs a vehicle, manufactures it and sells it once. Revenue is tied largely to unit sales. The company may earn additional money from financing, servicing or software, but the main transaction still happens when ownership changes hands. A robotaxi model changes that completely. Instead of generating revenue one time at the point of sale, the same vehicle could potentially generate revenue over and over again by selling rides throughout the day.

That is why autonomous ride-hailing is such an important idea for Tesla. A privately owned vehicle often spends most of its life parked. A robotaxi, in theory, becomes a productive asset. If it can operate for many hours a day, carrying passenger after passenger, the same car begins looking less like a consumer product and more like infrastructure. The financial value of the vehicle no longer comes only from what someone is willing to pay to own it. It comes from how much transportation revenue the vehicle can produce over time.

This is a very different business model from traditional car manufacturing, and it pushes Tesla closer to something that resembles a hybrid of automaker, software company and transportation platform. The company is no longer just asking how many vehicles it can sell. It is asking how many rides each vehicle can complete, how efficiently the fleet can operate and how much demand exists for driverless transportation. That moves the conversation from hardware margins to utilization, network density and recurring revenue.

The appeal of that model is obvious. A company that successfully operates autonomous vehicles at scale could capture much more lifetime value from each car than a one-time sale would provide. It could also potentially reduce reliance on the normal replacement cycle in which customers buy a new vehicle only every several years. In that sense, the most valuable transformation may not be making a better car. It may be turning the car into a machine that continuously earns money.

But that vision also explains why the path is difficult. Building a robotaxi business involves much more than manufacturing the vehicle itself. The company must prove the safety of the technology, satisfy regulators, manage public trust, secure operating permits, build the ride-hailing system and maintain the vehicles as part of an active fleet. A traditional carmaker mainly needs to persuade a customer to buy the car. A robotaxi operator must persuade cities, regulators and the public to accept an entirely different way of moving through everyday life.

There is also a broader lesson here for other industries. Some of the most powerful business transformations happen when a company stops earning money only when the product is sold and starts earning money from what the product does after it is deployed. Software shifted from one-time licenses to subscriptions. Industrial equipment increasingly includes ongoing monitoring and service contracts. Media moved from individual purchases to recurring access. Tesla’s robotaxi push reflects the same logic in physical form: the biggest opportunity may be turning an owned product into a recurring service.

The Cybercab therefore represents more than an unusual vehicle design. It is a visible example of a much larger economic shift. If autonomous transportation becomes mainstream, the winning company may not simply be the one that builds the most cars. It may be the one that best turns those cars into revenue-generating assets inside a functioning network. At that point, the question is no longer just whether a customer wants to buy the vehicle. It is whether the vehicle itself has become the business.