The Coming Age of Agentic AI

To combine with one of the previous articles?

The Coming Age of Agentic AI

Recent advances in generative Artificial Intelligence (AI), using Large Language Models (LLMs) based on the Transformer architecture, have led to significant improvements in natural language chatbots, as exemplified by ChatGPT. At the same time, these LLMs are being increasingly applied towards the development of Agentic AI — AI systems that are designed to perform tasks autonomously, make decisions, or take actions on behalf of users or other entities. Unlike chatbots, which can only answer questions, agents can perform tasks on behalf of their users. This trend captures both the development of dedicated agents and the gradual agentification of existing chatbots, such as ChatGPT.

Early toy examples of dedicated AI agents included Baby AGI and Auto-GPT, which while interesting experiments were unable to solve all but the simplest of tasks. Over time, more advanced and capable agents have been introduced, including those built using the OpenAI Assistants API and Custom GPTs, the Rabbit R1 hardware-based mobile agent, or bespoke agents and agent networks built with developer frameworks like Langchain and Microsoft’s Autogen. These advances reflect several key trends which are becoming increasing apparent.

4 Emerging Trends

1. Rapidly Improving Capabilities: As the capabilites of the underlying LLMs continue to improve, alongside our understanding of how to harness them through agentic software architectures, the capabilites of AI are increasing multiplicatively (cite paper). Narrow agents are giving way to truly generalist agents, which are beginning to look like Samantha, the personal OS imagined in the popular movie Her.

2. Widespread Adoption of Agents: As the usefulness of AI agents continue to improve they will quickly become widely adopted. While ChatGPT was the fastest app in history to reach 100 million users, when OpenAI later launched GPTs (agents) over three million had already been developed at luanch. Rabbit R1 has already sold through six batches of pre-orders and is poised to replace the smartphone as our everyday personal device.

3. Increasing Reliance on Agents: As we delegate more and more responsibility to agents they will become an integral part of our daily lives. They will act as our personal digital assistants and trusted companions. They will mediate our interaction with the digital world, and by extension, our interaction with other humans online. To do so, they will need access to our online identities, accounts, and private credentials.

4. Agents will become the UI: As agents become more capable, widespread, and integrated into everything we do, they will become the standard User Interface (UI) for all software, replacing traditional graphical user interfaces such as mobile apps, web browsers, and desktop operating systems.

Agentic AI Safety

For these reasons, AI experts are becoming increasingly concerned about the risks associated with mass adoption of AI agents. These concerns have led to early proposals to define a governance framework for Agentic AI.