Deep Dive: Exploring AI Agents for Enterprises
You've probably seen the onslaught of AI related content and determined either been intrigued or you're sick of hearing about it. It's also possible to be in both camps as well - I know I am. There is a lot of content being generated online, but it doesn't cover the depth of knowledge you should demand. Most of these are just scratching the surface.
I've been supporting organizations with digital transformations for over a decade and projects at scale over $1 Billion. In this next phase of autonomous transformations, the realm of possibilities is significant. I want to start a series that explores from an enterprise architecture perspective
What is security, privacy, and regulatory concerns for such a rollout?
What's possible with agentic models? How to setup checks and balances?
Biases that you need to be aware of in models, and how to work with them
Managing infrastructure & cost for such a deployment
Can this be done in a sustainable manner (ie. minimize carbon footprint)?
I've decided the best way to do this is start my own project and share the findings as I build out agents and understand the implications of orchestration for these agents. As it's bleeding edge technology, I don't have all the answers today but I'm excited to explore these avenues.
Why you should care about AI
Before we dive deeper into the project I'm looking to build, let's touch base on why you should care about what's happening in AI. Periods such as the Industrial Revolution, and the Digital Age have had significant impacts economic, social, global, political, and environmental impacts.
We're now about to enter the era of AI (or at least that's the vision that's being sold to us) - does that mean Artificial General Intelligence (AGI) is around the corner (ie. AI that can reason like humans and consider multiple diverse domains in it's reasoning)? I suspect not. I think there's lots of interesting use cases prior to AGI that are worth exploring.
Exploring new use cases
I've decided to start a moonshot project that will help me distinguish reality from fiction, and I will share my lessons with you.
I read a significant amount of content from case studies to rss feeds from various sources. It's a common underlying need to all my work - from consultancy to even writing on here. The problem I face is there is a lot of noise - simply put poor quality content with little substance.
How do I determine if it's something worth reading? In my case, I'm looking for data analysis, sources, relevancy, timeliness, etc.
At one point I was considering hiring a researcher who could help me decipher what's worth my attention. For the purposes of this experiment, I'm going to try using an AI agent and see if it can support my needs.
Setting Project Parameters
I know OpenAI has DeepResearch as an offering, but it's entirely private. The issue here is we don't know if our data is safe, given the fact that their mission has changed, and the organization has made many questionable choices (remember they stole Scarlett Johansson's voice without her approval), I would like to avoid them.
The parameters & needs for this project then are:
the LLM is open sourced
Running on my infrastructure (this will be cloud)
Cost-effective, secure, and minimal carbon footprint
Evaluating output
I will be judging the success of this project against the 5 questions asked above. Additionally, I will be reviewing the following items:
Scalability to meet demanding workloads
Performance metrics to determine latency in real-world scenarios
Opportunities for integrations
Continuous learning mechanisms for the agent - how will it adopt new patterns over time?
Upgradability of selected solution
Compliance and Ethics - all models are biased by those who train them, we will need to understand that
Cybersecurity analysis against zero trust principles
Let me know if have any suggestions for additional parameters, I would be interested to hear from you.