Earlier this month OpenAI made headlines when it introduced “Swarm,” an experimental framework designed to coordinate networks of AI agents. Though not an official product, Swarm provides developers with a blueprint for creating AI systems capable of autonomous collaboration on complex tasks.
What most people don’t realize is that teams of AI agents- known as Agentic AI- already have the ability to take over entire departments, replacing – and outpacing -white collar workers, says Simon Torrance, the London-based founder and CEO of two advisory firms: ‘AI Risk’ – a research, strategy and innovation network focused on developing and deploying AI effectively and safely -and ‘Embedded Finance & Insurance Strategies’ , which helps leaders across multiple sectors leverage fintech and insurtech to create new value and growth.
Torrance cites the following example: A year ago the entire operations team of a small European insurance brokerage left the company after being poached by a competitor. The CEO asked a friend, a successful insurtech entrepreneur and computer scientist who had just sold his predictive analytics business for advice. The friend- who prefers to remain anonymous for now – had some spare time to help and decided to try something radically new: re-designing the whole operations function using AI alone, with tools that already existed and were readily available.
While the company used some freelancers and other staff as stop gap measures the friend analyzed all the job tasks and workflows and, within three months, created a new ‘team’ comprised exclusively of AI agents that took on the roles of a commercial manager, an actuarial and underwriting function, an accountant, customer care managers and IT staff for the insurance brokerage. The brokerage’s data was not prepared for AI but many of these roles and processes are quite generic and Large Language Models (LLMs) could replicate them quite easily, says Torrance. That said, the friend in charge of the project had to very carefully map out the processes and then use other human colleagues and LLMs to enhance and develop them.
Once they were unleashed the Agentic AI outperformed its claims ratio objective by a factor of two, says Torrance.
Operating expenses plus claims costs typically determine net profit in the insurance industry, and human salaries, benefits and payroll taxes often make up around 65% of total operating expenses for a broker. By replacing humans and improving the claims ratio the AI team reduced those costs for the European brokerage to zero, says Torrance.
Typically, insurance underwriting generates a net profit (premiums minus claims costs and operating expenses) of around 5% in a good year. The Agentic AI team helped to generate insurance net profits of roughly 45% which was “completely unheard of and also unethical,” says Torrance, so tweaks had to be made.
Ethical, legal and safety guardrails clearly need to be put into place, but the European brokerage’s results give a glimpse of what is possible now, he says. The potential upside for enterprise is so great that Torrance and the computer scientist that built the Agentic agents for the insurance brokerage are in the process of creating a company that will offer corporates a no code product to create and manage teams of digital workers that are capable of autonomous decision-making, executing complex tasks with minimal human intervention, collaborating with other systems and people and dynamically adapting to changing conditions.
“Companies will be able to increase profitability by hiring an almost infinite number of ‘assistants’ and even workers at relatively zero cost” says Torrance.
For example, recently, Nvidia CEO Jensen Huang shared his vision of a future where, by roughly 2030, his company employs 50,000 human workers (up from 30,000 today) aided by 100 million ‘AI assistants’, notes Torrance. “In terms of ‘AI agents or ‘digital co-workers’ undertaking knowledge worker tasks end-to-end either independently or as parts of hybrid human-digital teams I would say that Nvidia could increase its ‘workforce’ (humans + digital) by 50%. So, by 2030, it might have 50,000 human workers, 25,000 digital workers (autonomous agentic workers) and 100 million AI assistants. That’s for a super tech company. Normal companies could perhaps increase their ‘headcount’ by 10 or 20%, without incurring the costs associated with human employees.”
“The digital Agentic workers are often better than human workers because they are not constrained by human biases about what’s possible or not, and when you give them a task they can keep going, they don’t need to take time to take lunch, drive home or sleep,” says Torrance. “What is key here is that we are not talking about single agents doing one thing. These workers are not assistants for humans. Agentic AI is made up of a diverse set of workers that can collaborate to get things done.”
Agentic AI will have a big impact on enterprise because it directly impacts three of the foundational pillars of a corporate’s competitive advantage: operational efficiency, scalability, and agility in decision-making, says Torrance.
Gartner predicts by 2028, 33% of enterprise software applications will include Agentic AI, up from less than 1% in 2024, enabling 15% of day-to-day work decisions to be made autonomously. The firm has named it a top strategic technology trend in 2025.
Yes, But Is The World Ready For Digital Workers?
Capgemini reports that 10% of large enterprises are already using AI agents, more than half plan to use them in the next year and 82% will adopt them within the next three years.
Yet when Torrance posted about the insurance brokerage example on LinkedIn earlier this month, some people insisted he had made the whole thing up.
“I can’t help feeling this is some kind of thought experiment that is on its way to being concretized into an urban myth,” wrote one person commenting on LinkedIn. Torrance insists the story about the European insurance brokerage is real, but says he is not permitted to reveal the name of the brokerage. The Innovator was not able to independently verify the example.
Other cases involving huge efficiency gains when AI agents replace humans have already come to light. As Sebastian Siemiatkowski, the CEO of Swedish fintech company Klarna told his shareholders recently: “Our [customer] AI assistant now performs the work of 700 employees, reducing the average resolution time from eleven minutes to just two, while maintaining the same customer satisfaction scores as human agents”.
People are skeptical because they are not ready for it to be true, says Torrance. He points to what happened to Lattice, a human resources and performance management platform that offers performance coaching, talent reviews, on-boarding automation, compensation management and a host of other HR tools to more than 5,000 organizations around the world. On July 9, the company said that it would begin to support digital employees as part of its platform and treat them like any other employee, only to be greeted by a swift and brutal backlash.
“Today Lattice is making AI history,” Lattice CEO Sarah Franklin pronounced on LinkedIn. “We will be the first to give digital workers official employee records in Lattice. Digital workers will be securely onboarded, trained and assigned goals, performance metrics, appropriate systems access and even a manager. Just as any person would be.”
According to Franklin, it’s avatars like Devin the engineer, Harvey the lawyer, Einstein the service agent and Piper the sales agent who have “entered the workforce and become our colleagues”. These AI agents have already been introduced by companies like customer Salesforce and startups like Cognition.ai and Qualified to perform work in lieu of humans. Salesforce’s Einstein, for example, can help sales and marketing professionals predict revenues, complete tasks and liaise with prospects. Cognition’s software engineer Devin can plan and execute complex engineering tasks requiring thousands of decisions, while recalling relevant context at every step as it learns over time, and fixes its own mistakes. Qualified’s sales rep Piper “works around the clock to convert inbound website traffic into pipeline” and is “bright, hard-working, and crushes her pipeline targets”.
Franklin was tapping into what promises to be the future of recruitment. Bringing AI agents into the enterprise is “going to be more like on-boarding employees than writing software,” Huang, the CEO of Nvidia, said during a fireside chat with Salesforce CEO Marc Benioff at the company’s annual Dreamforce conference in September.
Some humans do not take kindly to this news as Lattice quickly discovered after posting its plans on LinkedIn last July.
“This strategy and messaging misses the mark in a big way, and I say that as someone building an AI company,” said Sawyer Middeleer, an executive at a firm that uses AI to help with sales research, said in a reaction to Lattice’s post on LinkedIn. “Treating AI agents as employees disrespects the humanity of your real employees. Worse, it implies that you view humans simply as ‘resources’ to be optimized and measured against machines. It’s the exact opposite of a work environment designed to elevate the people who contribute to it.
The Guardian reported that the backlash was enough to force Franklin to suspend her company’s plans three days after her announcement.
The Next AI Wave Is Coming, Like It Or Not
The AI revolution can be broken down into waves, says Salesforce, which in September announced that it will be making a “hard pivot” in its approach to AI with Agentforce – an artificial intelligence platform that allows users to build and deploy autonomous AI-powered agents on Salesforce’s existing apps.
The first wave was predictive AI where AI models forecast the future based on patterns or historical data. For example, predictive AI can suggest the next best action for a sales professional to take with a prospect based on previous engagement. Such technology is also common in streaming services that monitor subscribers’ viewing habits to recommend other content they might like.
The second wave was generative AI. Powered by Large Language Models (LLMs) which can understand and generate humanlike text, generative AI models can create new content — from marketing copy and campaign images to sales emails and customer service. This wave led to the creation of consumer bots like ChatGPT. While automated in nature, they still require human input, through prompts, to get a job done.
Now, we are entering the third wave where autonomous AI agents, or Agentic systems, not only recommend actions but can reason and tackle multi-faceted projects without requiring human oversight at every step, says Salesforce.
The American Customer Relationship Management giant foresees a future in which AI agents will interact with each other to drive even greater efficiencies.
“Right now, we’re seeing little dollops of AI productivity being added to existing processes, but over time what we expect is that work will truly be transformed,” Mick Costigan, VP of Salesforce Futures, a team that explores how technology might shape business and the world, is quoted as saying in a Salesforce blog post. “We’ll have agents monitor each other. We’re already seeing in some of the early research that the ability for agents to work in a network and to have one agent whose job it is to check in on the actions of another agent actually improves the outcomes.”
AI Agents That Slack, Sulk and Exceed Expectations
In the example of the European insurance brokerage the computer scientist became the only ‘human in the loop’, providing the artificial operations team with their macro-objectives, monitoring progress, ‘coaching’ the AI agents to do better, and fixing problems.
Each AI agent was given a human name, and some acted in strangely human ways; sulking, for example, if their ideas were rejected or the wrong tone was used in communications, says Torrance. But the biggest problem with the new artificial team was that it was too good at its job.
The insurance industry – today – is based on the principle of mutualization or ‘risk pooling’: diverse groups of people pay into a pot, managed by an insurer, which allows those who are unlucky enough to incur losses from an accident or incident to be financially compensated, explains Torrance. The risks are spread and shared across this population – some of the customers (bad drivers for example) are more likely to claim than others (good drivers). By including both types in the pool, insurers can spread the costs of insuring everyone, setting affordable premiums that work for the group as a whole. That’s what regulators oversee.
Agentic AI potentially disrupts this centuries-old industry model, which has assumed a world of partial and poor quality data and a limited ability to process it. AI is now able to undertake – and act upon – more dynamic and real-time risk assessments in ways that shifts the calculation of risk probability from the group to the individual level. Mass personalization and new types of risk management solutions become increasingly feasible.
“The insurance brokerage’s AI agents proved too effective at reducing the claims costs: they used their relatively infinite cognitive abilities to over-achieve on the objectives set for them – reducing the claims ratio and keeping customers happy,” says Torrance. “By analyzing data in smart new ways, they could identify certain types of customers who were more likely to make claims, at certain times or in certain situations. “
The AI agents rapidly designed and activated interventions that reduced the frequency of these situations: rejecting certain customers, urging them to cancel subscriptions at certain times in their lifecycle, personalizing pricing and adjusting terms and conditions. All of this was technically legal in their jurisdiction but, for a company in the insurance industry, it wasn’t ethical, says Torrance.
As a result, the European insurance brokerage had to get its computer scientist friend to quickly train the AI agents on ethical principles that form the basis of the company’s values. “This lead to much less impressive financial results, but would ensure more people were covered by insurance (a societal good) and avoid raising red flags with the regulator,” says Torrance. Having successfully solved the CEO’s problem of the operations department, the company is now looking to use Agentic AI in sales and other functions, he said.
A Gigantic Opportunity
There is a reason why Salesforce is betting its future on AI agents. It is a gigantic opportunity, Benoiff said during his Dreamforce keynote speech. Benoiff spoke about how Agentic AI can give businesses that are over stretched needed capacity. “What if these workforces had no limits at all?” he asked. “It is a strange thought but a big one.”
Companies need to let that sink in, says Torrance. “Suddenly there is a new potential for a new set of (digital) co-workers that are almost infinite that can augment what employees are doing, replace traditional roles or do something completely new,” says Torrance. “People don’t realize that this is technically possible today.”
AI agency is a spectrum, says Gartner. At one end are traditional systems with limited ability to perform specific tasks under defined conditions. At the other end are future agentic AI systems with full ability to learn from their environment, make decisions and perform tasks independently. “A big gap exists between current LLM-based assistants and full-fledged AI agents, but this gap will close as we learn how to build, govern and trust Agentic AI solutions,” says the research group.
Hyperscalers like Microsoft and Amazon are already adding agentic AI to their AI assistants.
For example, in October Microsoft announced a suite of autonomous AI agents for its Dynamics 365 platform, intensifying the competition with Salesforce in the enterprise AI market. The tech giant said it will release ten new autonomous agents designed to augment sales, service, finance, and supply chain teams. Available in public preview starting next month, these AI agents aim to automate complex tasks and orchestrate business processes across organizations, surpassing traditional chatbots and Microsoft’s earlier AI offerings by reasoning over intent and context, making judgments based on a broader set of data.
Startups also see an opportunity. This week Anthropic, the Amazon-backed AI startup founded by former OpenAI research executives, announced AI agents that can use a computer to complete complex tasks like a human would. Meanwhile, VentureBeat reported that CrewAI, which was founded in 2023 and competes with startups like LanGraph and AutoGen, launched CrewEnterprise, a new platform built on top of its open-source framework which it said enables organizations to build crews of AI agents using any large language model (LLM) or Cloud platform. Users can plan and build multi-agent systems; securely deploy those agents into a production environment with custom levels of access and control; and iterate and track return on investment (ROI) with testing and training tools.
Managing The Risks Of Agentic AI
Along with opportunity Agentic AI brings challenges that require advanced tools and strict guardrails, says Gartner. It lists proliferation without governance or tracking; bad output due to reliance on low-quality data, employee resistance and Agentic AI driven cyberattacks enabling smart malware among the risks.
To get started with Agentic AI Gartner recommends:
- Look for agentic AI in your technology stack. Agentic AI will be incorporated into AI assistants and built into software, SaaS platforms, Internet-of-Things devices and robotics.
- Leverage APIs and events to enable agentic AI. This will allow AI agents to interact seamlessly with various tools and environments, ensuring they can execute tasks and receive information effectively.
The tools are here but CEOs have a dilemma, says Torrance. “Which parts of the organization do they augment and empower and which do they automate and replace? And in what order, and how fast?”
To get started he recommends the C-Suite analyze every job in an organization and put a dollar value on how much it would cost to augment every single task. “Once you have done the analysis you can determine what does this mean for corporate strategy,” he says. “Ask yourselves what could be done if we had 1000 additional workers in a particular department?” Don’t wait, says Torrance. “Agentic AI is a future that is already here.”
This article is content that would normally only be available to subscribers. Sign up for a four-week free trial to see what you have been missing.
To access more of The Innovator’s Focus On AI articles click here.