Initial Thoughts on Agentic AI & It's Impact

My Exposure to Agentic AI

I try to frequent many different conferences and events to learn more from the source and to expand my network. One of my main interests, especially as of late, has been AI, and with the recent industry craze about Agentic AI I've gotten to see some hands on demos and experiments. Note: the things I write in this article are based on conversations and interactions I've had with contributors to the AI field, and it is important to remember with a fast evolving technology the applications mentioned today may or may not be relevant in the future.

What is Agentic AI?

Agentic AI is an artificial intelligence system designed to complete a task with minimal supervision. It is built of individual agents that are powered by LLMs acting as the “brain”. The idea is they work together to automate tasks and workflows. See a definition from IBM.

I was discussing the topic with a group of people from varied backgrounds, and someone who actively is working on developing an agentic model for contract resolution had a unique way of framing the trust their company puts into the technology. They described Agentic AI to be around the same level as a "junior" position. In the case of her program, it was a "junior" level output for finding potentially flawed clauses in a contract. I think that for the application she was trying to automate it made proper sense to hold the system to that level, but I think that different systems should be held to higher standards and others more laxed. It really comes down to the importance & impact of the data the system interacts with.

Effect on the Job Market

The biggest question surrounding the AI Field in general is "Will this take my job?" Overall out of the current AI tools and workflows, Agentic systems are most likely to replace human workers. The decision really comes down to a few key factors: ROI, function, trust, and creativity.

• ROI: to put it simply, is it more cost effective to have a team of humans working on a project or service if an Agentic ystem could do the same for a fraction of the cost. This also goes the other way for many different services; in which, it would actually be more costly and time consuming to rely on the Agentic System.
• Function: there are currently many tasks that AI and Agentic systems can still not complete due to the necessity of human reasoning. This also applies especially to physical based jobs in which the robots are still being researched and tested. In the next couple years, it will be crucial that we pay attention to physical automation not just "White-collar" jobs.
• Trust: data security and potential leaks are the biggest concern above function. Many of the popularly used LLM's have already had major security risks exposed. This must be examined in the scope of Agentic aswell. Especially under the current position to view them as "junior" level positions. Think, "If this was a junior in [BLANK] position, what information and permissions would they have?" This is also very interesting to examine for A2A registries/protocols. There are concerns on how AI implementations can disrupt internal Data Hierarchy (Grouparoo's explantion) which is based off Maslow's hierarchy of needs.
• Creativity: Agentic systems are really good at handling logic based tasks but when it comes to actual media production it is still based on LLM's. There are definelty some good uses from a "creative" prospective such as having it write emails etc. but when it comes to image or video generation there is a large gap in quality. However, this lack of qualitity is not stopping many individuals. Across social media are automated n8n and similair workflows producing hundreds of posts, videos, and adverts. This means the "replace-ability" for creative positions comes down to quality compared to quantiity.

Final Remarks

Regardless of opinions on AI and automation, it is important to understand the tools that are starting to become more common. I think that we may hit a limit and the AI bubble will slow down similair to the IoT bubble in the early 2000's. This was a very crude overview, so make sure to be on the lookout for other posts with more indepth articles.