AI Agents Struggle with Basic Business Tasks, Highlighting Need for Human Judgment, Says Ben Eubanks

In a recent LinkedIn post, Ben Eubanks sheds light on the current limitations of artificial intelligence in handling everyday business operations, drawing from a study by Carnegie Mellon University. Eubanks details an experiment where AI agents were tasked with common workplace duties, revealing significant shortcomings that underscore the continued importance of human oversight and judgment in the business world.

AI Agents Fall Short in Simulated Business Environment

The experiment involved a simulated company staffed entirely by AI agents. These agents were assigned tasks such as sending emails, updating records, and managing workflows. However, the results, as described by Eubanks, were far from seamless. Many AI agents struggled with fundamental operations, with the top performer completing only a quarter of the assigned tasks. This performance level is particularly striking given that the tasks were designed to be akin to those a minimally trained human intern could handle.

Eubanks highlighted several critical failures observed in the AI agents:

• Many agents were unable to even begin tasks, getting stuck on basic steps like navigating file systems.
• Document handling proved problematic, with some agents overwriting files or failing to adhere to required formatting.
• Social and administrative tasks also presented challenges, with one agent misunderstanding a user-update workflow and causing structural damage.
• A concerning number of agents entered infinite loops or produced confident but incorrect actions without self-correction.

As Ben Eubanks notes, these issues are not rare edge cases but reflect the current reality of AI capabilities:

“None of these are edge cases. These are everyday office tasks a summer intern could do with almost no training.”

Human Judgment Remains the Differentiator

The shortcomings of the AI agents in the Carnegie Mellon experiment, as pointed out by Eubanks, lead to a crucial conclusion: the technology is not yet ready for full autonomy in business operations. Eubanks argues that despite the rapid advancements and hype surrounding AI, it lacks the essential elements of grounding, judgment, and contextual understanding that human employees possess.

According to Ben Eubanks, human intuition plays a vital role in navigating the complexities of the workplace. Humans can identify when instructions seem contradictory, when a workflow needs flexibility, or when the tone of communication is inappropriate – abilities that current AI systems largely lack.

“Human judgment remains the differentiator. Humans know when a file path “looks wrong,” when instructions conflict, when a workflow needs to adapt, or when a communication tone doesn’t fit the situation. AI still doesn’t.”

Augmentation, Not Substitution: The Path Forward

Eubanks proposes that the most effective approach to integrating AI into the workplace is through augmentation, rather than outright substitution of human workers. He observes that successful teams are not aiming to replace people but are instead focusing on how to combine human strengths with AI efficiency.

In this model, AI can manage repetitive and data-intensive tasks, freeing up human employees to focus on higher-level responsibilities. As Ben Eubanks points out, this synergy allows for better outcomes:

“The teams getting the best results aren’t trying to eliminate people; they’re pairing human strengths with AI efficiency. AI handles the repetitive parts. Humans steer and make the decisions that matter.”

This shift also places a greater emphasis on the development of uniquely human skills. Eubanks suggests that as AI takes over routine tasks, the value of human creativity, problem-solving abilities, communication skills, and adaptability will only increase, fostering greater organizational agility.

A Collaborative Future for Work

Concluding his post, Ben Eubanks frames the current moment as a pivotal one for the future of work. AI is powerful enough to significantly alter how businesses operate, yet it remains unreliable enough that human involvement is indispensable. He challenges the prevailing question of replacing humans with AI, suggesting a more productive inquiry:

“Instead of asking “How do we replace humans?”, a better question is “How do humans and AI work together to create better outcomes?”

This perspective underscores the idea that the most promising future involves a collaborative relationship between humans and AI, leveraging the distinct advantages of each to achieve superior results.