Generative AI copilots could promise ‘a workplace utopia’

“We’re not just trying to provide information and resources and the latest and greatest things just for those that are in the office,” says Lam.

Jamf, which helps enterprise customers manage Apple devices, operates a fully hybrid workplace, and in some smaller markets, it would be costly to hire support staff. That led Lam to invest in Moveworks’ artificial intelligence copilot, integrated in Slack, and provide automated support for functions like IT and HR.

Over 70% of Jamf’s workforce is using the Moveworks copilot today, known internally as Caspernicus. Lam says the copilot helps Jamf automate some work and helps employees be more productive as they don’t always have to wait for an IT colleague to get back to them when a problem arises. But she’s also wary of perpetuating tech sprawl. “We don’t want to end up in a situation where we are doing five or six different copilots,” says Lam. 

Copilots promise a workplace utopia, making employees more productive, improving workflows, and helping share knowledge across an organization. One study found that software developers using Microsoft’s GitHut Copilot were able to complete tasks 56% faster than those not using the tool.

During Microsoft’s latest earnings call in April, the tech giant said GitHub Copilot had 1.8 million paid subscribers, up over 35% from the prior quarter. Nearly 60% of Fortune 500 companies use Copilot for Microsoft 365, with a client roster that includes Nvidia and Amgen.

But at a cost of $30 per user for Copilot for Microsoft 365, some companies have been cautious about how many employees they are willing to sign up to the service. 

For rivals looking to compete with Microsoft, Boston Consulting Group partner John Pineda says, niche copilot offerings need to go beyond the underlying power of the models they sit on and focus more on domain specificity that reflects decades of servicing a specific field, like healthcare.

“The onus is on the more specialized players,” says Pineda. “If it looks a lot like your Microsoft Copilot, I think it’s going to be a lot harder.”

Thomson Reuters has a plugin with Microsoft’s Copilot, a combination that allows users to work in Microsoft Word but have access to Reuters’ expertise in more dense topics including taxes and law. 

“Copilot on its own is not designed to answer legal questions or help with tax-related work,” says Kriti Sharma, chief product officer of legal tech at Thomson Reuters. “Our partnership with Microsoft is to bring us intelligence right where the users work.”

For businesses that operate outside Microsoft’s ecosystem, Thomson Reuters also sells a generative AI assistant called CoCounsel, which can handle research, drafting, and document review. “The use case for our CoCounsel experience is how we can speed that up, give them more confidence in the answers, and most importantly cite the sources, which is really important for legal work,” says Sharma.

In May, software provider ServiceNow and Microsoft announced the integration of two generative AI assistants, ServiceNow Now Assist and Microsoft Copilot, allowing a user in Microsoft Teams to ask workplace-related questions to Now Assist. The integration will be sold as a premium ServiceNow offering beginning this fall. Future capabilities will include allowing employees to create documents, like a PowerPoint presentation, based on ServiceNow prompts.

“It’s just a series of features that get enabled inside of your existing ServiceNow applications that use generative AI,” says Pat Casey, chief technology officer and a cofounder at ServiceNow. “The idea that we would need an external entity to coordinate AI services and interact with the underlying application, that’s just not the approach we took.”

Last year, consulting giant Accenture was one of the first companies to have access to Copilot for Microsoft 365, Security Copilot, and GitHub Copilot. Arnab Chakraborty, chief responsible AI officer at Accenture, says that whenever a copilot is built—either in house or by a third-party vendor—an easy feedback loop needs to be established so that those testing the technology during the pilot phase can share problems that need to be addressed.

Chakraborty says guardrails are also needed to ensure employee personas are established to ensure that when a copilot shares a response to an inquiry, it isn’t giving every employee access to information that may be more restrictive.

And then there’s training. As copilots roll out across organizations, Chakraborty says workers are getting access to a new tool they’ve either never used before or have had little past experience with. “We’re giving them a guidebook,” says Chakraborty, of Accenture’s workforce. “That goes a long way to have the right discipline in using those agents.”

SAP launched a generative AI copilot, Joule, last fall; since then, the enterprise software company has horizontally expanded the assistant to support HR tool SuccessFactors, SAP’s cloud offering, and by the end of this year, will be fully integrated within all major SAP applications. “We see great uptick by customers,” says SAP chief AI officer Philip Herzig.

Integrating Joule horizontally will allow a manager to look up an employee’s compensation, which would be found in a HR application; but then, if the next question is about if there’s room in the budget to give this employee a raise, that’s a financial question. The intent is that rather than requiring a manager to switch applications, Joule sits in both and can assist with questions across multiple departments.

“In order to make it a really conversational experience that follows what the user wants, this integrated approach is key,” says Herzig.

Unstructured, a startup that ingests and pre-processes unstructured data that can be ready for use with foundational models, has leaned on kapa.ai’s chatbot to help answer questions from the 3,000 engineers that are part of a free Slack community created by Unstructured.

The chatbot is cheap to operate and helps answer user questions across the globe. Unstructured CEO and cofounder Brian Raymond says success of the chatbot is based on the company providing feedback to the chatbot and then making adjustments to the documentation.

When it comes to copilots, AI chatbots, or AI agents, Raymond says Unstructured had been “very skeptical of the efficacy of it. We’ve actually grown more confident over time.” 

But he says success is still narrow. The technology is most adept at digesting a question and providing an answer. Raymond is more skeptical of complex, multistep automation that’s being demoed by copilot vendors.

“The reality is the tech just needs quite a bit more maturing in order to do it,” Raymond says.