Agentic AI: The top challenges and how to overcome them

Reliability and predictability

The way we interact with computers today is predictable. For example, when we create software systems, an engineer sits and writes code, telling the computer what to do step by step. When using the agent-based AI process, we do not provide step-by-step instructions. Rather, we provide the outcome we want to achieve, and the agent determines how to achieve that goal. The software agent has a certain degree of autonomy, which means that the output may be random.

We have seen a similar problem with ChatGPT and other LLM based applications. generative AI systems when they first debuted. But over the past two years, we’ve seen significant improvements in the consistency of generative AI results thanks to fine-tuning, human feedback loops, and consistent efforts to train and improve these models. We will need to make similar efforts to minimize the randomness of agent-based AI systems to make them more predictable and reliable.

Privacy and Data Security

Some companies are hesitant to use agent-based AI because of privacy and security concerns, which are similar to, but can be even more concerning than, generative AI. For example, when a user interacts with large language modelevery bit of information passed to a model becomes embedded in that model. There is no way to go back and ask him to “forget” this information. Some types of security attacks, such as rapid injection, exploit this by attempting to trick the model into revealing sensitive information. Because software agents have access to many different systems with a high level of autonomy, there is an increased risk that they may expose private data from more sources.