High-risk, high-reward partnerships powering Anthropic's AI expansion

Anthropic which has aggressively pushed into enterprise business, announced new updates to its tool Claude Cowork that extend its abilities to areas like human resources, investment banking and design

By Parmy Olson

 

Anthropic PBC made an announcement last week that flew under the radar. Having sparked selloffs in the equity market with a relentless stream of product updates, the company wanted to avoid rattling investors more. Its livestream wasn’t widely announced; there was no accompanying blog post. But the launch was about as meaningful as anything the artificial intelligence lab has lobbed into the market recently.  

The company, which has aggressively pushed into the enterprise business with “agents” that carry out tasks, announced new updates to its tool Claude Cowork that extend its abilities to areas like human resources, investment banking and design. To do this, Anthropic worked with several companies, including FactSet Research Systems Inc., S&P Global Inc. and PriceWaterhouseCoopers LLP, to make those capabilities possible through so-called plug-ins. 

 

 

A plug-in is like a mini-app inside of Claude Cowork that augments the AI tool with a new skill, such as financial analysis. FactSet, for instance, co-developed three plug-ins for Claude Cowork, according to Anthropic, which do things like draft research notes, compare consensus estimates and update financial models.

 

This is great for Claude — but what’s in it for FactSet or these other third parties? Anthropic only makes a vague suggestion that being plugged into its AI will make partner products stickier. “The more connected Claude is to an organisation's existing software, the more useful it becomes — and the more valuable those tools become in turn,” Anthropic says in its press release. But that still doesn’t make financial sense for Anthropic’s partners who are providing expertise, essentially through step-by-step instructions, that allow an AI tool to replicate some of what they do, whether it’s how equity analysts compile corporate reports or human-resources staff create performance summaries. 

 

The motivation for companies like FactSet and Thomson Reuters Corp., another partner, is simple: If Claude is going to reshape how their industry works, they’d rather be inside the machine than outside watching it all happen.  

 

But for now, at least, the arrangement has dark echoes of the partnerships that Facebook struck with publishers about a decade ago, where news articles were freely shared with the social network in return for more traffic. When Facebook changed its algorithm in 2018 to deprioritise news, all the new traffic vanished, and publishers were left with a failed investment and ad businesses that had withered in the meantime. 

 

Anthropic says its arrangement is nothing like Facebook’s, adding that its plug-ins offer “bidirectional value” and make its partners’ products more useful and stickier. It also doesn’t cost anything to provide a plug-in for Cowork. Still, the measurable return for participants who provide their expertise is unclear, especially when Anthropic is capturing all the financial benefit through subscriptions to Claude Cowork that cost tens or hundreds of dollars a month, depending on usage, per “seat” for corporate teams. 

 

Let’s give Anthropic the benefit of the doubt for a moment. What if it were laying the groundwork for a bigger future financial opportunity for its partners by, say, sharing revenue? That wouldn’t be unheard of in this new business of AI models; OpenAI, for example, gets a cut from whatever Microsoft Corp. makes selling subscriptions to Copilot, since it provides that product’s underlying AI system. 

 

A better and more ambitious strategy for Anthropic would be create a marketplace where partners earn a commission when their apps are used. That would create a financial incentive that attracts more partners and potentially more subscribers, not unlike the network effect that Apple Inc. triggered when it launched its App Store in 2008. It would diversify Anthropic’s revenue beyond just subscription fees. 

 

It would also be a healthy step toward filling in what tech media publisher Tim O’Reilly calls “the missing half of the AI economy.” Driven by Silicon Valley’s blitzscaling mantra and competitive dynamics, top AI labs have focused on monetising models solely for themselves through things like subscriptions or selling API tokens. The partners helping to make Claude more valuable are getting little in return.

 

“Market-making is the best way to justify all those capital expenses,” O’Reilly tells me, referring to the billions earmarked for chips and data centres by AI firms. He points to how Alphabet Inc.’s Google originally developed pay-per-click advertising, while YouTube used ContentID (a tool for detecting copyrighted music) and revenue sharing to resolve disputes with the music industry. Both cases led to new ways for others to earn money. No major AI lab has done that yet. 

 

Anthropic has become the company of the moment for how it has shaken markets, and for standing up to the US Department of Defence over potentially unsavoury uses of its tech. The latter has given the company a noble sheen of late, but exploiting the expertise of other companies could do the opposite, especially if it contributes to the white-collar jobs bloodbath that Anthropic’s CEO predicts.

 

It may be sensible to remain sceptical about all of these disruptive possibilities. Anthropic’s plug-ins may flounder, just as OpenAI’s “GPT Store” did in 2024. Sam Altman’s marketplace for custom-built ChatGPT agents never gained traction thanks to quality control issues and — guess what? — a lack of opportunities for partners to make money. Perhaps that’s now something to reconsider.   (Disclaimer: This is a Bloomberg Opinion piece, and these are the personal opinions of the writer. They do not reflect the views of www.business-standard.com or the Business Standard newspaper)