While it may be hard to anticipate all the reverberations that autonomous work agents may set off, a few discrete use cases are already emerging. — Photo by Cash Macanaya on Unsplash
If you’re just getting up to speed on chatbots and copilots, you’re already falling behind. Talk in Silicon Valley now is squarely focused on agents – artificial intelligence (AI) that can handle multistep chores like onboarding clients, approving expenses and not just routing but actually responding to customer service requests, all with minimal human supervision.
OpenAI Chief Executive Officer Sam Altman calls agents "the next giant breakthrough.” Salesforce Inc. has already signed deals to install AI agents at more than 200 companies including Accenture Plc, Adecco Group, FedEx Corp., International Business Machines Corp. and RBC Wealth Management.
"We're really at the edge of a revolutionary transformation,” Salesforce CEO Marc Benioff said on the software company’s most recent earnings call. "This is really the rise of digital labour.”
Sceptics may note the similar tone of excitement that accompanied the 2022 debut of ChatGPT. While the OpenAI chatbot dazzled, it has yet to widely unlock substantial productivity gains or radically alter most workplaces. Agent technology goes a step further, not just performing parlour tricks and spitting out plausible responses to queries but actually doing the kinds of repetitive tasks that today are handled by millions of humans.
Agents are not just for workplaces. You might use one someday soon to research, select, and book each component of a lengthy vacation itinerary, for example. What Altman finds more exciting, though, is the prospect of an agent that acts like "a really smart senior coworker who you can collaborate on a project with,” he said in a podcast interview last month. "The agent can go do a two-day task – or two-week task – really well, and ping you when it has questions, but come back to you with a great work product.”
While it may be hard to anticipate all the reverberations that autonomous work agents may set off, a few discrete use cases are already emerging.
Salesforce launched its agent product, which it calls "Agentforce,” in September. Anthropic launched its own product in October, followed by a Microsoft launch in November. OpenAI is set to unveil an agent at a research preview in January. Meanwhile, a slew of startups have launched their own "agentic” AI products.
Patrick Stokes, executive vice president for product and industries marketing at Salesforce, said in an interview that he expects to see agents becoming more widespread as soon as the first quarter of 2025.
"November of next year, we'll be looking back at this like, ‘How was there ever a world where this didn't exist?’” Stokes said.
‘Roles will shift’
At a recent Salesforce event in New York, executives were sure to underscore that agents aren’t meant to steal jobs. "It’s not going to replace, it’s going to augment,” Saks Global CEO Marc Metrick said in a Salesforce promotional video that played during the keynote.
But, as Benioff said on the earnings call, changes will need to be made, including at Salesforce. "The transformation is not without challenges, jobs are going to evolve, roles are going to shift and businesses will need to adapt,” Benioff said. "We're all going to need to rebalance our workforces as agents take on more of the workforce and then we can rebalance and reshape our companies into new ways.”
At consulting firm McKinsey & Co., an AI agent now handles the tedium of client onboarding. It coordinates paperwork, shares relevant contact details, affirms the scope of the project – and runs everything by the firm’s legal, risk, finance, staffing and other departments to get their signoffs.
"That used to be an absolute spaghetti bowl of email threads between all the different functions,” said Rodney Zemmel, who leads McKinsey’s digital practice and the firm’s own AI transformation. In the past, onboarding required tens of hours per new client. Now, "an agent basically does all that chasing for you,” and completes the process in roughly 30% of the time. It sends emails and follows up to wrangle whatever information it needs to move projects forward. The final product is then reviewed and approved by a human.
"It works in this case because it's a complicated set of tasks, but actually a fairly standardised and routinised one, without too much judgment involved,” Zemmel said.
Another application McKinsey is testing is a "squad” of agents to work together like a team of human employees would. McKinsey often helps companies migrate data from mainframes to the cloud – a "laborious, complicated, expensive process,” Zemmel said. So McKinsey trained an agent squad to mimic the different team members that would typically staff the project, like designers and data engineers.
While the squads are not yet fully operational, Zemmel said initial results have been impressive. "You can cut the time to do a mainframe to cloud migration more than in half by using these agent squads with the right degree of human supervision over the top,” he said.
Agents can be activated with natural-language instructions and are designed to be equally conversant with their human users. McKinsey recently gave access to an internal platform that lets everyone in the firm build their own agents. Allowing each employee to design the agents that would be most useful to them could bring major productivity benefits. But also, Zemmel said, "the potential to create absolute chaos is there,” so building the right safeguards around the agents is essential. At McKinsey, a central team will review all agents against cyber, risk, legal and data policies before they’re made available to the rest of the firm.
Companies that adopt agents won’t replace entire departments overnight, Zemmel said. The new technology may even lead to a reversal of offshoring for functions like human resources, finance or tech. Instead of outsourcing work to a large team with an average skill set in a country with lower labour costs like India, companies get more out of employing a small but highly skilled team at home that’s leveraged by powerful agents.
Fooling even mom
Agents are versatile in that they can be strictly for internal use or set up to "speak” with clients. Nsure, an online insurance company, deployed an AI agent that communicates with customers via phone, text, email and online chat, answering questions, providing quotes, logging information and solving problems.
The agent, named for Nsure’s vice president of AI and automation, John Haisch, handles 60% of customer requests, taking over repetitive, time-consuming tasks that were previously shouldered by human employees. "Friendly John,” as the agent is known, was trained on 300 hours of Haisch’s voice, which resulted in a simulacrum so convincing that the executive’s own mother thought it was him, just with a slight cold.
Nsure also trained the agent on insurance law and regulatory information, plus three years of human agent interactions with customers, with the conversations annotated to correct any shortcomings. And the agent is continually improving as the company feeds its most recent conversations back into the model to show it anything that resulted in a transfer to a human agent.
Nsure co-founder and Chief Operating Officer Wojtek Gudaszewski said that with AI agents, the company can grow the business with only 20% of the workforce it otherwise would have needed to achieve the same scale. But it hasn’t downsized. With the work set up like a production line, it simply moved human agents into a different place on the line – from roles that required more repetitive tasks to ones that are more complex.
This shift led to higher morale and lower turnover, Gudaszewski said, with employees placed into higher-paying, meaningful roles where they can learn and grow. Another perk: Since the AI agent can handle customer inquiries all hours of the day, human employees don’t have to worry about working nights or holidays.
So far, the experience at Nsure tracks with the hypothesis laid out by Dario Amodei, co-founder of AI company Anthropic. He suggested the arrival of AI agents could translate into higher pay for humans with skills that AI can’t replicate.
"In fact, even if AI can do 100% of things better than humans, but it remains inefficient or expensive at some tasks, or if the resource inputs to humans and AI’s are meaningfully different, then the logic of comparative advantage (for humans) continues to apply,” Amodei wrote in a wide-ranging meditation on how he thinks AI will shape our future.
"However, I do think in the long run AI will become so broadly effective and so cheap that this will no longer apply,” he wrote. "At that point our current economic setup will no longer make sense, and there will be a need for a broader societal conversation about how the economy should be organised.”
Different agent types
Long before we potentially reach that point, companies will have tested a wide range of applications for AI agents.
At Accenture, plans are underway to build industry-specific agents for clients across sectors, from travel to retail banking to supply-chain planning.
Internally, the company has already deployed agents to help its marketing department project the return on investment for events or craft a marketing plan based on historical data and information about what competitors are doing. The agents were first piloted in August and now have been rolled out to about 500 marketing employees.
Most of the agents are "utility” agents that function like a junior researcher would. But the company has also built what it calls "strategic” agents, which can coordinate the work of multiple research agents, similar to a team leader. The agents also can "huddle” amongst themselves, sharing information like employees would in regular check-in meetings.
Lan Guan, Accenture’s chief AI officer, said the strategic agents can work side-by-side with junior marketers and even act as coaches for them. In the past, a junior employee tasked with running an event might get nervous at the thought of having to write a marketing brief or do the strategic planning. Overwhelmed, they might seek the advice of a supervisor. "That’s the old way of working,” Guan said.
Now, she said, a strategic agent will have the context and memory to understand what the employee is trying to do and can orchestrate the right utility agents, based on past experience and best practices, to get the task completed.
Accenture declined to say if its hiring projections have changed with the efficiency gains expected from using AI agents. – Bloomberg
