.gif)
Imagine you need to insure your new car. You call an insurance broker to discuss the type of vehicle, whether it's new or second-hand, how many kilometers you intend to drive per year, who will be driving the car, the type of coverage you want, and based on this, the insurance options available to you — except the voice on the other end of the phone is an AI agent.
Today, this is possible, and it raises an important question: is generic AI sufficient to meet the specialized needs of various professions, or do we need dedicated AI agents tailored to each one? At Hexa, we’re betting on the latter.
An AI agent is a system that makes decisions and takes actions, similar to a human. It can make phone calls, book appointments, process payments, and do other basic tasks, across many fields. It’s designed to mimic human actions, learning and improving with every interaction.
Needless to say, its potential to boost company productivity is immense.
However, generic (1) agents face limitations. AI, in general, can’t reason like humans. As AI pioneer Yann LeCun highlights, current AI systems lack the fundamental ability to think and adapt in situations in the nuanced way humans do. If they haven’t encountered the situation before, they can’t intuit an appropriate response. This presents a problem for generic AI agents, whose tasks span multiple domains, each with unique requirements and contexts. Training a generic AI to excel in every domain would necessitate an impractical amount of data and resources. So, these agents often end up being "jacks of all trades, masters of none," unable to deliver high-level performance in any specific area.
The challenges faced by AI agents are clearly seen when we look at their application within complex fields. Take autonomous driving as an example: companies like Tesla and Cruise have trained their AI agents on millions of hours of driving data, yet their systems are still far from being autonomous, due to the sheer amount of data needed to be able to react appropriately to every possible situation. A human, on the other hand, can achieve basic driving competence with just 20 hours of instruction.
The case for specialized AI agents
Specializing AI agents allows them to become highly proficient in specific domains. They can leverage integrations with specialized software and be trained on large amounts of highly specialized data, ‘limiting’ the actions they are designed to perform, enhancing their ability to deliver complex tasks with greater accuracy.
Take our car insurance broker as an example. A generic AI assistant might handle basic tasks like appointment scheduling and customer reminders, but a specialized car insurance broker AI agent could do much more. It could manage your customer records, coordinate with you to identify its VIN, check different alternatives based on insurance contract options, order a vehicle registration plate, and even get it delivered to your address — in short, it could do the job of a person working in a car insurance brokerage.
According to Sarah Tavel, this is where the opportunity lies: in selling the work itself rather than just software. This shift means offering AI solutions that deliver complete work products, directly competing with human labor. In our car repair example, a specialized AI agent takes over substantial portions of the workflow, delivering completed tasks and allowing garage employees to focus on higher-level duties. It’s about creating a whole extra headcount.
Our vision
At Hexa, we believe the current opportunity in AI agents lies in identifying specific functions within professions that can be taken over by specialized agents. By focusing on well-defined tasks, we can develop AI solutions that deliver superior performance and capture significant market share. This strategy mirrors the success of Vertical SaaS Software, where hyper-specialization has proven to be highly effective.
Distinguishing between AI copilots and agents offers additional opportunities — in the former field, we’ve just launched Tandem. While agents take over tasks entirely, copilots serve as intelligent assistants, augmenting human capabilities without fully replacing them. Both approaches have their strengths, and the choice depends on the specific needs of the business and its clients.
The debate is open! Whether you're interested in developing vertical AI agents, exploring the copilot model, or anything else AI-related, we’d love to hear from you.
Florent Quinti, partner at Hexa AI, is looking for founders for his next (specialized!) AI project - reach out if our thesis resontes.
(1) A Generic Agent is capable of handling a wide range of tasks, such as answering phone calls, booking meetings, and updating CRM records, across various industries. This versatility mirrors the expectations of a human agent, who must be able to handle multiple types of inquiries and tasks.
In contrast, a Specialized Agent is dedicated to tasks within a specific field, such as serving as a phone assistant in the car repair industry. This focused expertise allows for more in-depth knowledge and efficiency in that particular area.
.webp)
.webp)