The Morning Download: Long-Running AI Agents Are Here

Good morning. A $300 billion selloff in software and data stocks on Tuesday suggests that the dreaded bubble may not be what lies within the AI sector, but what exists beyond.

Investors have been concerned for months that AI could suck the value out of narrowly focused software-as-a-service companies, rendering them mere databases that feed AI agents. It isn’t difficult to see how a new generation of long-running AI agents could end  up stepping between the customer relationships that SaaS companies have cultivated, much as Apple extracted the value in mobile communications from hardware-focused incumbents that didn’t get software.

Those fears were compounded in November by the growing power of Anthropic’s Claude Code, which builds software with stunning proficiency.

On Tuesday morning, investors homed in on Anthropic’s announcement that it was adding new legal tools to its Cowork assistant meant to help automate a number of legal drafting and research tasks.

It was sufficient to give them a glimpse into the afterworld of economic life as we know it.

Co-What?

Cowork is a variation of Anthropic’s Claude Code agent, both of which are based on the company’s Claude Opus 4.5 model released in November. Claude Code was quickly adopted beyond the world of developers, and the company launched Cowork in January with a graphical user interface optimized for broader tasks beyond coding.

As a result, investors who spent last year debating the existence of an AI bubble were already starting to take AI’s potential a lot more seriously, given its rapid, recent breakthroughs. The Cowork upgrades announced on Tuesday were like a match on dry tinder. If AI agents like Cowork can disrupt one research-intensive field such as the law, why not another and another? And suddenly this seemed like a near-term threat, not a speculative problem that’s always five years in the future.

I spoke to Scott White, Anthropic’s head of product for enterprise, to get a better understanding of how these models work and how they stand to transform work, companies and markets. 

Every business leader needs to understand these technologies, given their rapid development and broad, potential impact.

Going long

Claude Code and Cowork are pioneers of an emerging class of long-running AI agents with implications far beyond any one particular sector such as software development.

Unlike previous iterations that generated responses immediately, Claude Opus 4.5 creates a budget of effort for every parameter, assessing the complexity of a requested task and determining how hard it needs to think to solve it, White told me recently.

That’s how the model can engage in "long-horizon" thinking, he told me. The fusion of a reasoning model with a capable product engine or "harness" allows the model to connect to real systems, execute code, and manage workflows, he said. Users can now assign a broad objective to the AI, which will operate autonomously in the background, maintaining context over long horizons. The agent can work for longer periods of time, but the turnaround time for projects is much, much faster.

Changing work

White said these new capabilities are already changing the way he and his colleagues work.

In the past, he wrote documents to convince engineers to build features. Now he uses Claude to build working prototypes to demonstrate viability. The near-instantaneous merging of internal data, web research, and customer feedback compresses weeks of analysis into minutes of execution.

Anthropic designers now write production code to implement their own interfaces, and product managers are performing complex data science tasks that previously required specialists, he said.

Work is changing at three levels, he said. Individuals are becoming more capable and productive and companies are tearing down historical workflows from marketing to compliance, reducing turnaround times.

“Lastly, businesses are changing how they think about what they're going to look like, what products they want to build,” White said. “Now that building new things is so much more approachable for businesses, introducing new revenue lines.”

She mastered a snowboard trick over thousands of runs. Ahead of the Olympics, AI convinced her to change it. The WSJ Leadership Institute's Isabelle Bousquette reports how AI developed by Google DeepMind is giving athletes like Maddie Mastro precise, 3-D insights into how their bodies move.

The snowboarder is one of a handful of U.S. Ski and Snowboard athletes testing the new AI tool that leverages advanced computer-vision models, purpose built to understand the human pose tracked through 3-D space over time.

Mastro tells Isabelle how the tool, after analyzing video footage of her crippler, a signature jump trick, flagged that her lead arm was angled too high on landing, a flaw that could cost points as she goes for Olympic gold in Italy this month.

“I would have never noticed that if it didn’t highlight it and show me my body positioning,” she said.

It's the "subtle adjustments," that make all the difference between first place and missing the podium, three-time Olympic halfpipe gold medalist Shaun White, tells Isabelle. White also advised on the AI tool’s development

Companies are often using only about 10% of the Copilot seats they pay for, according to Citi Research, which flagged “disorganized data silos” as a persistent issue.

Microsoft has several versions of Copilot that are woven into apps and services, including its 365 productivity tools such as PowerPoint, and the GitHub developer platform. There is also a consumer-facing version available through its Edge browser and via an app.

Who’s flying the plane? Employees say an organizational silo between Microsoft AI CEO Mustafa Suleyman’s consumer team and the enterprise-facing groups has made a unified product vision difficult.

Model development has also been constrained by limited computing capacity, forcing Microsoft to ration server time to serve OpenAI and other Azure customers.

And adding pressure, Anthropic’s new Claude Cowork (see above) has drawn praise for working fluidly across 365 apps, a flow many Copilot users have struggled to achieve.

11.5%. The percentage of Microsoft Copilot subscribers who used the product as a primary option during the July through late January period, down from 18.8%, according to a survey by market research firm Recon Analytics.

15 million. The number of corporate Copilot “seats” in Microsoft's 365 productivity business, which has a base of 450 million-plus paid seats overall.

$1 trillion. The market capitalization Walmart hit Tuesday, a first for a retailer not named Amazon.com. Walmart’s stock has surged in recent months, fueled in part by Wall Street enthusiasm for the company’s online business as well as its investments in automation and AI.

AMD sales climb. Advanced Micro Devices reported a 34% jump in fourth-quarter revenue to $10.27 billion, beating expectations as its data-center business surged 39%, the WSJ reports. The company expects revenue from its data-center unit to increase by more than 60% in the next three to five years as AI hyperscalers’ demand grows, CEO Lisa Su told analysts.

HP CEO jumps to Paypal. HP Chief Executive Enrique Lores is leaving to take over PayPal, replacing CEO Alex Chriss in a surprise move that left the PC maker scrambling. HP named board member Bruce Broussard as interim CEO. Lores also chairs PayPal’s board, a role that includes evaluating CEO candidates, Semafor notes.

Another high-profile exit hits Salesforce.Tableau chief, Ryan Aytay, has left the company, adding to executive churn at the business software maker, Bloomberg reports. Recent exits include Slack CEO Denise Dresser, now OpenAI's chief revenue officer, and cyber chief Brad Arkin. Analysts tell Bloomberg that the turnover raises questions about Salesforce’s health. Shares are down 42% over the past year.