Co-pilots, move out of the way – time to make room for our AI agent.
That’s the message coming from the software industry in recent days, with several major companies backing the latest ideas to turn generative artificial intelligence into a work commodity.
Microsoft, Salesforce and Workday this week put Agent at the center of their AI plans, while Oracle and ServiceNow also touted the idea at the industry’s annual user conference this month.
AI assistants called “Copilot” (a term first popularized by Microsoft) have become the software industry’s main response to the generative AI unleashed by the launch of ChatGPT about two years ago.
Modern AI agents are going even further, designed to act on behalf of their users. They have become the front line in a new battle between tech giants like OpenAI and Google, while also marking the software industry’s latest attempt to sell generative AI to business customers.
This evolution reflects both advances in the underlying technology and new marketing strategies from an industry seeking to capitalize on much-hyped technology that has yet to have a significant impact on revenues.
If the industry’s claims prove true, the transition from AI assistants to agents could open the door to a much more disruptive phase in the evolution of generative AI, both for the workers affected by the technology and the software companies themselves.
The rise of agents (also broadly referred to as “agent” systems) stems from a number of advances in underlying technology since the first generative AI chatbots.
Larger memory allows the system to better understand context and improves planning capabilities. Agents often connect to other systems through APIs (application programming interfaces) and can take actions on behalf of the user rather than simply returning information.
The latest generation of agents is designed to act as an extension of the original Copilot, rather than replacing it entirely: Microsoft CEO Satya Nadella said the company’s Copilot software is evolving into an “enterprise orchestration layer,” a conversational interface that lets employees create agents to perform specific tasks.
Recommendation
Initially, AI agents were primarily touted as tools to take over simple, everyday actions, like filling out expense reports.
But some companies are already touting their ability to handle more complex tasks or even take over some jobs entirely. Automating customer support systems has become a major focus, potentially replacing many call-center employees.
So far, generative AI has done little to boost revenue growth for software companies.
Jim Tierney, a growth investor at AllianceBernstein, said the software industry as a whole is still in “‘Show me, show me’ mode when it comes to co-pilots and AI agents.” “How this gets monetized is still an open question,” he added.
Salesforce CEO Marc Benioff suggested Copilot’s support was lacking, telling the Financial Times: “Microsoft has fooled customers with its AI strategy. Customers don’t need to build AI themselves. We’re building AI into our platform. Customers shouldn’t be forced to train models over and over again.”
Software companies believe that customers will see direct productivity gains from having agents who can take on entire tasks.
According to Microsoft’s Nadella, as these AI systems become more capable, “the models themselves will become more commoditized, and all of the value will come from building on, manipulating and tweaking these models to align with your business data and workflows.”
As agents take on more work, companies like Apple, with its dominance on the smartphone platform, and Microsoft, with its desktop productivity apps, are more likely to win, Tierney said.
For now, the overall impact of that change is likely to be modest, as the tendency of generative AI systems to “hallucinate” will make users wary of allowing the AI to act without oversight.
Barry Briggs, a former chief technology officer at Microsoft who is now an analyst at independent research firm Directions on Microsoft, said he is “skeptical, even a little bit worried” about widespread use of the agent.
He added that the probabilistic nature of the technology means customers can’t use it for critical tasks, but instead need to embed it into work processes that give workers the final decision-making power.
But some companies claim they’re already taking the technology to its logical conclusion: Last month, Sebastian Śmiatkowski, CEO of Swedish fintech firm Klarna, said the company plans to use AI to cut its headcount in half.
Siemiatkowski also shocked the software industry by saying Klarna would abandon Salesforce and Workday entirely and instead use AI to develop the software businesses need to run their businesses — a claim widely considered heresy in the tech industry given the current tech landscape, but which some argue hints at a far more disruptive future.
But most software investors are betting that the industry’s incumbents will be the big winners, even though it’s still unclear when and how the technology will pay off.
As an extension of their co-pilot, agents are just the latest step in today’s big software companies’ efforts to secure their territory and prepare for a time when generative AI has advanced to the point where it can deliver real productivity gains, said Kevin Walkusch, a portfolio manager at Jensen Investment Management.
While the current generation of agents is unlikely to contribute significantly to software companies’ revenues, “the key is to establish a foothold and long-term positioning,” he added.
Even if incumbents are likely to win, the shift to agent-based systems could wreak havoc on the way they do business: Most companies charge customers license fees based on the number of employees who use their software, but this model would be threatened if AI agents were to seriously impact employee numbers.
In response, most software companies have started testing usage-based pricing, which ties revenue to query volume. For example, Salesforce has announced that it will charge $2 per “conversation” with its AI agents. Many companies are also talking about moving to outcome-based pricing, where they would share a portion of the profits that customers make from using the software, but it is unclear how this will work.
“It’s still early days, so it’s hard to say how the pricing model will evolve,” said Byron Dieter, a partner at Bessemer Venture Partners and an investor in early-stage software companies.
Just as changes to how software companies book revenue, like the move to cloud, have caused a period of disruption in the industry, the transition to new pricing models for AI “could be challenging for public[software]companies,” he added.
