AI Agents have been making a ton of headlines over the past few weeks, but what exactly are AI agents and how should we be thinking about them? It’s helpful to start by going back to AI Assistants, which were all the rage before AI agents took over. Let's explore the upgrades that transformed AI Assistants into the AI Agents we have today.
When ChatGPT launched in 2022, it took the world by storm. Its conversational style, ability to understand user intent, and creative outputs were revolutionary. However, there were clear limitations in this early AI assistant framework that have since been addressed, leading us to the development of AI agents.
One of the first improvements was incorporating better instructions and direction for AI assistants. This process, known as prompt engineering or custom instructions, involves being clear about what we expect from these AI assistants, including the personas they should assume and behaviors to avoid.
Another significant breakthrough was enabling AI assistants to reference documents before responding to a user. Initially, AI assistants could confidently produce incorrect outputs and lacked domain-specific or company knowledge. Now, with access to source material, they can incorporate relevant information into their responses, reducing errors and increasing expertise in specific fields.
Originally, ChatGPT had a gap in its training dataset, meaning it couldn’t provide information on recent events. By integrating search capabilities, AI assistants can now look up and incorporate current information, ensuring they stay relevant and up-to-date.
AI assistants have also gained access to tools like calculators and programming environments. Large language models are inherently text-based, so adding mathematical engines like Wolfram Alpha or dedicated compute environments allows them to handle complex tasks more effectively.
Using advanced prompt engineering and orchestration techniques, AI assistants can now plan steps in advance to solve more complex problems. They also have enhanced troubleshooting capabilities, allowing them to manage more complicated projects seamlessly.
A major upgrade is the ability of AI assistants to interact with external systems. They can now pull data from business systems and push tasks to external systems for completion, significantly expanding their utility.
Recent advancements have enabled AI assistants to interact with users beyond just text. Users can now upload images and, soon, leverage cameras and video, opening up a range of new use cases.
These advancements have culminated in the creation of AI agents, which are better equipped to handle complex problems. They can plan ahead, troubleshoot effectively, use advanced tools, and interact with users in innovative ways. AI agents also perform actions in the real world by integrating with other systems.
Despite their promise, AI agents present a few challenges:
The AI community is excited about AI agents, and major players like OpenAI and Google are launching their versions. Companies like Ushur are developing AI agents tailored for customer experience automation in sensitive industries such as healthcare, insurance, and financial services.
