09-07-2025
What are AI agents and which jobs are they coming for first?
For months, Silicon Valley has raved about the potential for AI agents to transform the way we work and live. While much of the public's attention has been glued to generative AI programs such as ChatGPT, major AI companies are locked in a tight race to refine large language models to the point that they have the capability to execute tasks on their own. In a January 2025 blog entry, OpenAI chief executive Sam Altman predicted that this will be the year when 'the first AI agents 'join the workforce' and materially change the output of companies.' But what exactly is an AI agent and how likely is it that they will not just assist but replace human workers? The Financial Post's Yvonne Lau explains.
An AI agent is an application of a sub-category of artificial intelligence technology known as 'agentic AI,' which refers to AI systems that are autonomous. Like other AI systems, agentic AI collects and processes data to perform a function. Current generative AI programs — the likes of ChatGPT and Midjourney — create new content ranging from text to images and strings of software code using predictive modelling. Agentic AI's key difference lies in its agency (hence its name): it can manage complex tasks and make decisions on its own, based on pre-set user goals. These are tasks that involve the ability to reason, solve problems and be adaptable in response to changing situations.
Gary Filan, a partner and head of AI at KPMG Canada, said the definition of AI Agent is rapidly evolving, but that a common way to think of it is as a 'virtual assistant or coworker.'
AI agents can be deployed across a wide spectrum of industries and job functions. Some early successes have come in the fields of customer service, human resources, market analysis, and fraud detection and prevention, to name a few.
'Financial services firms are the leading group in that list,' Filan said. These businesses have used AI agents, for example, to streamline insurance claims processing, generate risk assessments, to help answer customer queries, among more, according to KPMG Canada.
For example, a run-of-the-mill AI assistant could help a bank send fraud alerts to its customers. But an AI agent could take it one step further: it can oversee real-time transactions, flag suspicious activity and work with the bank's fraud detection system to prevent malicious activity.
Manufacturing firms have leveraged AI agents to monitor equipment performance, predict failures, and dispatch maintenance teams before downtime can occur, while retailers have used them to predict demand trends, adjust stock levels, and fine-tune pricing, said Shannon Katschilo, the Canada country manager for Snowflake Inc., an AI and data cloud platform.
Rote job functions, like those in the areas of customer support, analytics, engineering and field operations, are likely to be among the most affected, Katschilo said.
But both Katschilo and Filan argue that AI agents won't necessarily replace jobs or workers. 'It's the nature of the work (that) will change,' Filan said. 'Workers may switch from a rote processing type of role to one that involves more judgment, monitoring, and management — and ability to work with these AI systems.'
Most Canadian organizations have now heard of agentic AI: almost three-quarters said they were 'very familiar' with the concept, according to an April 2025 KPMG Canada survey of 252 business leaders.
Still, uptake remains limited in Canada even while the majority of participants — 88 per cent — said that adopting agentic AI will help their organization become more competitive.
Only 27 per cent of respondents in the same survey said that their organizations have adopted or deployed agentic AI and have active use cases in their organization, while 57 per cent plan to invest in, or adopt, agentic AI in the next six months.
A second quarter 2025 Statistics Canada report on AI use by businesses in Canada found that only 12 per cent of them reported using AI to produce goods or deliver services, though that figure increased from six per cent during the same time last year.
'Canadian firms are on the laggard side with regards to adoption of agentic frameworks. The primary reason is Canadian corporations' aversion to risk,' Filan said. But Canada is now experiencing a shift with the Carney government's focus on so-called 'light and tight' AI and technology regulation, he said.
At their core, AI agents are built on large language models (LLMs) that sometimes 'hallucinate' — meaning that they can generate outputs that range from nonsensical to simply false.
The autonomy of AI agents, alongside the growing sophistication of the technology, magnifies its risk factors. Agentic AI systems that operate independent of human oversight could carry out unintended and potentially harmful actions, from leaking confidential data, to impersonating people or manipulating files.
The risk also depends on how, and where, it is applied. Agentic AI applications for 'healthcare or human resources — to decide who gets a raise or who gets laid off, for example — are much more critical than agentic AI for a food delivery app,' said Mélissa M'Raidi-Kechichian, research and advocacy fellow at the Center for AI and Digital Policy, a Washington-based non-profit focused on AI best practices. 'What remains across contexts though, is the accountability component: without human oversight, who will be held accountable when an AI system inevitably fails, does harm, or does not perform the way it was intended to?'
Some in the industry are already working toward establishing guardrails to advance the development of safe and ethical AI. Canadian-French computer scientist Yoshua Bengio, considered one of the 'godfathers of AI,' recently launched a Montreal-based non-profit called LawZero focused on AI systems that will filter out certain traits like dishonesty. He aims to create a tool to de-risk AI agents and keep them in line. 'I'm deeply concerned by the behaviours that unrestrained agentic AI systems are already beginning to exhibit — especially tendencies toward self-preservation and deception,' Bengio wrote in a June 2025 blog post.
Agentic AI technology is nascent, but developing rapidly, Filan said. 'I'm not even thinking about what it would look like 10 years from now. Most of the conversations occurring now are between a two-to-five-year period,' he said.
A growing number of startups are now developing AI agents customized for different professional and personal needs. Partners at Silicon Valley's fabled startup accelerator Y Combinator LLC, recently said that they have been bombarded with a wide range of AI agent proposals in fields ranging from marketing to recruitment and debt collection.
Silicon Valley leaders have warned that job displacement is coming rapidly. Anthropic PBC chief executive Dario Amodei told Axios in May 2025 that he thought AI could eliminate half of entry-level white-collar jobs and push unemployment to 10 to 20 per cent in the next one to five years.
Others say agentic AI technology has a ways to go.
Competition Bureau to probe whether Amazon's rules are unfair to sellers and consumers
Canadian AI darling Cohere to open Montreal office in rapid scale-up
A May 2025 Carnegie Mellon paper showed that Google LLC's Gemini 2.5 Pro, the top-performing AI agent, was unsuccessful 70 per cent of the time in completing real-world office tasks. Other rival agents created by tech giants like OpenAI and Meta Platforms Inc. had failure rates of over 90 per cent.
'Right now, we're seeing early glimpses: AI agents can already analyze data, predict trends, and automate workflows to some extent. But building AI agents that can autonomously handle complex decision-making will take more than just better algorithms. We'll need big leaps in contextual reasoning and testing for edge cases,' according to a March 2025 International Business Machines Corp. report titled AI Agents in 2025: Expectations vs. Reality.
• Email: ylau@
