AI must move beyond documenting tasks to truly automating work, as current agentic systems face significant limitations: industry AI expert.
An AI expert in a global research and business advisory firm has delivered a blunt assessment of the current state of agentic AI at a recent “Data & Analytics Summit” in Sydney, declaring, “AI is not doing its job today and should leave us alone.”
The industry expert criticized the proliferation of generative AI tools that summarize meetings and generate lists of action items, arguing that these systems merely add another layer of work rather than eliminating it.
His point was that AI should simplify users’ lives by automatically performing tedious tasks instead of just making it easier to document them. Highlighting an example from US healthcare company, he noted that employees had been polled about their most utterly-dreaded tasks. By automating these chores, the firm had seen immediate adoption and enthusiastic buy-in for further AI-driven automation.
Hyping up agentic nirvana
Another example involved a real estate firm that had used AI to automate a 17-step tenant screening process. Unfortunately, in practice, as the 17 steps were run in sequence, any failure at any point meant it the system had wasted time working on all steps prior to the failure. This business logic failure in programming AI was only solved when the system was set to run all steps in parallel.
Despite these successes, the expert, Erick Brethenoux, Global Chief of AI Research, Gartner, warned that the vision of fully autonomous AI agents — bots capable of independently performing complex tasks across an enterprise — is far from reality. While industrial firms have used AI agents in closed systems for decades, these agents struggle with complex, open-ended tasks. Tech vendors, he argued, are overselling the ease with which personal AI agents could manage multifaceted responsibilities, such as scheduling meetings while balancing the competing priorities of work and family life.
Brethenoux emphasized that building robust agentic AI systems is a significant software engineering challenge. It requires careful design around what agents can perceive, control, and execute, as well as how they communicate and negotiate with each other. He noted that vendors often gloss over these complexities, promoting a vision of “agentic nirvana” that is not yet achievable.
He concluded by criticizing the industry’s tendency to conflate “AI agent” with “generative AI”, warning that imprecise language only adds to confusion and hinders progress in developing truly useful AI systems.
