Artificial intelligence is usually framed as a tool employees use to work faster. Anthropic’s latest disclosure points to something more unusual. The company says Claude is now leading about 26% of the research and development work involved in building Anthropic’s next generation of AI models, up sharply from about 1% in March. Anthropic also says more than 90% of its research now involves some level of human-AI collaboration. That means the technology is no longer just helping people write emails, summarize documents or answer questions. It is increasingly helping engineers build the next version of the technology itself.

That creates a very different kind of productivity story. In most industries, a tool improves output by helping workers do the same job more efficiently. Here, the product is beginning to improve the process used to create the product. If each generation of AI can contribute more meaningfully to the design, testing and refinement of the next generation, development could start accelerating in a self-reinforcing loop. The more capable the system becomes, the more it may be able to contribute to making its successor even more capable.

Anthropic is trying to measure that shift more explicitly than most companies. It says the 26% figure was evaluated with help from Epoch AI, and it plans to publish the metric regularly so outsiders can track how involved AI is becoming in AI development itself. The company has also said about 30,000 AI agents are active on its internal research platform. In a sampled week in July, roughly 12% of compute used for AI-led research went to safety-related work, while only about 1 in 47,000 proposed actions by internal AI agents was blocked by safety screening. Those numbers suggest Anthropic is trying to show not only that AI is becoming more useful, but that it is doing so inside a structured system of human oversight.

That oversight matters because this is not the same as AI autonomously inventing its own successor without supervision. Anthropic says Claude does not operate independently and remains under human control. Engineers still guide the research, define the goals and review the results. But even with humans in charge, the nature of the work is changing. Researchers are increasingly managing, evaluating and directing AI contributions rather than doing every step manually themselves. In that sense, some of the most important jobs in AI may shift from creating every piece of work directly to designing workflows where humans and AI build together.

The business implications could be enormous. AI companies are locked in a capital-intensive race where speed matters. If one lab can shorten research cycles, test more ideas and improve models faster because its current system is helping create the next one, that becomes a competitive advantage. The value is not only in a better model. It is in a faster model-development machine. In the long run, that machine could matter even more than any single release because it determines how quickly a company can keep improving after competitors catch up.

There is a broader lesson here beyond AI labs. Many technologies create value by helping workers perform tasks more quickly. The bigger breakthroughs often happen when a technology starts improving the systems that produce more of that technology. Factories became more powerful when they started using machines to build better machines. Software became more scalable when programmers built tools that made writing software easier. AI may now be entering a similar phase, where part of its value comes from enhancing the process of AI creation itself.

This is why Anthropic’s announcement feels more important than a typical productivity update. It suggests the industry may be moving from AI as a helper to AI as a participant in its own advancement. If that percentage continues rising from 26% to something much larger, the pace of model development could begin compounding in a way that is very different from ordinary software improvement. The biggest productivity loop may begin when a technology starts improving the process used to build itself.

That does not automatically mean unlimited acceleration or the end of human control. It does mean that the frontier of AI competition may increasingly depend on who best designs the collaboration between people and machines. The companies that win may not simply build the smartest model. They may build the most effective system for using existing intelligence to create more intelligence. That is a very different business question and potentially a much bigger one.

Businesses spent the past few years asking what AI could do for them. Now, as AI agents begin handling real business tasks, a different question is becoming increasingly important: Who is watching the AI?

Enterprise companies are building systems to monitor AI agents, control what they can access and track the actions they take. Companies including Cisco, Intuit, Workday and ServiceNow are working on safeguards as autonomous AI becomes more deeply connected to business operations.

This is an important shift for businesses because an AI agent that can send an email, update a CRM, approve a workflow or interact with customers also has the ability to make mistakes at scale.

The implication for smaller businesses is straightforward: automation needs boundaries.

Companies adopting AI agents should know exactly what an agent is allowed to do, what information it can access, when a human must approve an action and how its activity can be reviewed afterward.

That doesn’t mean businesses should avoid AI agents. It means implementation is becoming more sophisticated. Connecting an AI agent to a company’s systems is only part of the job. Businesses also need permissions, monitoring and clear rules for what happens when the AI gets something wrong.

The next competitive advantage in business AI may therefore not simply be having more AI agents. It may be knowing how to deploy them without losing control of the business.

AI is becoming capable of doing more work. Now businesses have to build the infrastructure to supervise it.

For most of the social-media era, the deal was simple: Facebook, Instagram and WhatsApp were free, and Meta made money by selling advertising around the billions of people using them. Now Meta is testing whether that relationship can become something very different. The company has launched Meta One, a new subscription service offering enhanced AI capabilities and more than 50 additional features across Instagram, Facebook, WhatsApp and Meta AI. Meta says its phased rollout has already produced roughly 15 million subscriptions and trials, giving the company an early indication that at least some users may be willing to pay for a better version of something they have spent years receiving for free.

The opportunity is enormous because Meta does not have to start by finding customers. It already has them. Billions of people have existing accounts, established social graphs, years of photos and messages, favorite creators and daily habits built around Meta’s apps. Convincing even a relatively small percentage of those users to upgrade could create a significant new recurring-revenue stream without the customer-acquisition costs normally associated with launching a subscription business.

Meta One starts relatively cheaply, with individual app plans beginning at $2.99 per month and bundled individual plans starting at $7.99 per month. More advanced creator and business packages begin at $14.99 and rise substantially for professional users who want publishing tools, analytics, additional AI capacity, business messaging features and greater automation. Importantly, Meta says the core versions of its apps and Meta AI will remain free. The strategy is not to force everyone behind a paywall. It is to make the free product useful enough to attract billions of people while making the premium version valuable enough that some willingly upgrade.

AI could be the feature that finally makes that strategy work. Meta One subscribers can receive greater access to computationally expensive capabilities such as AI image and video generation, Instagram Restyle tools and other creative features. Meta says that in early testing, more than half of bundled subscribers used both AI and premium expression features, suggesting that people may not be paying for one single feature. They may be paying for a collection of small improvements that make an app they already use every day noticeably better.

That creates a very different business model from advertising alone. Advertising revenue depends heavily on how much attention users generate and how much advertisers are willing to pay to reach them. Subscription revenue comes directly from the customer and can be more predictable. Meta does not need subscriptions to replace advertising for the strategy to become valuable. Even a relatively modest subscription business layered on top of an enormous free audience could diversify revenue while helping offset the growing cost of providing advanced AI features.

There is a broader business lesson here. Companies frequently spend enormous amounts of money finding new customers while overlooking the people already using their products. Existing users already understand the product, have overcome the initial trust barrier and have developed habits around it. The challenge is no longer convincing them to try something unfamiliar. It is showing them that an upgraded version of something familiar is worth paying for. That can be a much easier sale.

Of course, Meta faces a difficult psychological hurdle because it helped train consumers to expect social networking for free. People may happily pay for music, movies or productivity software while resisting a monthly charge for Instagram features they once assumed should simply be included. That means the subscription must provide genuine recurring value. Custom icons and cosmetic features may attract some users, but AI creation tools, professional analytics, business automation and productivity features could be more important if Meta wants people to keep paying month after month.

Meta One therefore represents a fascinating experiment in monetizing an audience after the audience has already been built. Meta spent decades making its apps indispensable parts of everyday life and used advertising to finance that growth. Now AI gives the company a new category of expensive, valuable features it can place above the free tier. The biggest advantage Meta has may not be any individual AI tool. It is the fact that billions of potential customers are already inside the store. Sometimes the cheapest customer to acquire is the one using your product every single day.

Modern technology feels almost weightless. People talk about the cloud, artificial intelligence, digital payments and electric vehicles as if the future is built mostly from software. But the physical reality is very different. Every chip, battery, motor, server and electronic device begins with raw materials pulled from somewhere on Earth. That is why the U.S. government’s plan to begin issuing deep-sea mining permits within months matters. Washington is trying to secure access to critical minerals used in electronics, weapons and electric vehicles, and it is looking toward the ocean floor as a possible new source.

The idea sounds futuristic, but the business logic is very old. When an economy depends heavily on certain materials, the countries and companies that control those materials gain enormous leverage. Critical minerals are already central to batteries, semiconductors, magnets, advanced manufacturing and defense systems. As demand rises for AI infrastructure, data centers, electric vehicles and military technology, access to those materials becomes more valuable. Deep-sea mining is essentially an attempt to expand the supply chain before shortages, import dependence or geopolitical pressure become even bigger problems.

That is what makes this story bigger than mining alone. The most advanced technologies in the world still depend on extremely basic industrial foundations. Artificial intelligence may run on software, but that software needs servers. Servers need chips. Chips require highly specialized materials and manufacturing inputs. Electric vehicles may feel like a clean digital future, but they still require battery minerals, metals and large industrial supply chains. The cloud may look invisible from the user’s perspective, but the economy behind it is built from mines, refineries, factories and shipping networks.

Deep-sea mining highlights this hidden physical layer better than almost any other story. The ocean floor contains mineral-rich nodules and deposits that could potentially support industries looking for new sources of supply. If the U.S. starts permitting this activity, it signals that critical-mineral competition is becoming urgent enough that policymakers are willing to consider sources that once sounded too distant, expensive or controversial. In other words, the technology economy is expanding so quickly that it is starting to redraw the boundaries of where resource extraction may happen.

Of course, this does not make deep-sea mining simple. Environmental concerns are one major obstacle. The ocean floor is one of the least understood environments on Earth, and critics argue that large-scale extraction could damage ecosystems before scientists fully understand them. There are also political and legal questions around who gets access, how the resources are governed and how quickly any permits could translate into meaningful production. But the very fact that these questions are now being asked seriously shows how important critical minerals have become.

There is also a larger business lesson here. Companies often think of innovation as something that happens at the product level. A better battery. A faster chip. A smarter AI model. But some of the biggest competitive advantages are created much earlier in the chain. If a company or country secures access to scarce inputs before everyone else, it can strengthen its position across multiple future industries at once. That is why mining, processing and resource access are becoming strategic again after years of feeling like background infrastructure.

This pattern has happened before. Oil shaped the industrial economy because transportation, plastics and manufacturing all depended on it. Semiconductors became strategic because modern life depends on computing. Critical minerals are now moving into that category because they sit underneath so many technologies at the same time. The companies building the future may get the headlines, but the materials enabling that future can become just as important. The businesses supplying those materials may end up benefiting from almost every major technological trend at once.

The U.S. push toward deep-sea mining is ultimately a reminder that the digital economy is not as digital as it looks. Behind every elegant app, electric vehicle and AI model is a long chain of physical inputs that someone has to find, extract, process and move. Technology may feel increasingly abstract to consumers, but its foundation is still intensely industrial. The next generation of innovation may be powered by software, but it will still begin with digging valuable materials out of the ground or in this case, off the bottom of the ocean.

Lyft has begun offering fully driverless Waymo rides through its app in Nashville, and that move says something important about where the robotaxi business may actually create value. Years ago, Lyft spent heavily trying to build its own autonomous-driving technology through its Level 5 division. In 2021, it sold that division to Toyota’s Woven Planet for $550 million. At the time, that might have looked like Lyft stepping away from the future of self-driving transportation. Now it looks more like Lyft may have realized it did not need to build the robot at all.

That is because technology ownership is only one part of the business. Waymo can provide the autonomous vehicle, sensors, software and self-driving system. Lyft can provide something different but still extremely valuable: the customer relationship, the app, the marketplace, ride demand and the operational layer that connects riders to available vehicles. In other words, one company can own the machine while the other owns the flow of transactions around it.

This is a much bigger lesson than robotaxis alone. Businesses often assume they must own the core technology in order to win the market. Sometimes that is true. But in many industries, the winning position belongs to the company that controls distribution, customer access or the marketplace. Lyft may not manufacture vehicles, and it may not own the self-driving brain inside them, but it already knows how to attract riders, manage pickup and drop-off logistics and operate a transportation network in major cities.

That makes the economics of the robotaxi business more interesting. The obvious story is about autonomous-driving technology replacing human drivers. The more strategic story may be about which company controls the customer’s first tap. If someone opens Lyft first when they need a ride, Lyft remains central to the transaction even if another company supplies the actual vehicle. In many industries, the company that owns the customer relationship is often in a stronger position than the company providing the invisible technology underneath it.

This approach also allows Lyft to avoid one of the most expensive parts of the autonomous-vehicle race. Building self-driving technology requires huge amounts of capital, years of research, complicated testing and ongoing regulatory work. That is a difficult burden even for the largest technology companies. By partnering instead of building from scratch, Lyft may still participate in the upside of autonomous transportation without carrying the full research-and-development cost of inventing the technology itself.

There is a useful pattern here that shows up across business. Hotel-booking platforms do not need to own hotels. E-commerce marketplaces do not need to manufacture every product sold through them. Payment companies do not need to make the goods being purchased. In many cases, the business with the most durable position is the one that organizes demand, simplifies access and becomes the habit customers return to first. Lyft’s role in robotaxis may work the same way.

Of course, there are limits to this strategy. If the autonomous-vehicle provider becomes powerful enough, it may try to control the customer relationship directly. Waymo already has its own brand and its own presence in the market. That means Lyft’s long-term advantage depends on continuing to offer enough convenience, reach and rider loyalty to remain useful as a partner. Platform businesses are strong, but only as long as both sides still need the platform.

Still, Lyft’s move into driverless rides through a partnership shows a smarter and perhaps more realistic version of winning. The company may have concluded that it does not need to own the robot to benefit from the robotaxi future. It may only need to remain the place where riders go when they want transportation. In business, people often focus on who built the breakthrough technology. But the bigger winners are not always the inventors. Sometimes they are the companies that make sure customers use it.