Exploring AI Agent Frameworks: Can an Entire Business Be Run by AI Agents Alone?
Who enjoys working anyway?
Let’s see a show of hands…
OK, well, unless you’re maybe a professional athlete or a rock star, there probably aren’t too many of us that truly enjoy punching the clock in the outmoded “9-to-5” model anymore (if there ever was, that is). Yet, we all still have to pay our bills and taxes at minimum so it seems like we are destined to earn income in some capacity, right?
We can’t all be athletes or rock stars, so perhaps it’s time to hang up Guitar Hero. Furthermore, we don’t all want to be entrepreneurs or solopreneurs. Some of us just want to live our lives and have a money printing machine doing all the work for us. Sounds hard to believe and, yes, in most capacities it isn’t realistic. Unless you are born into wealth or got lucky to acquire money generating assets like real estate or businesses, most of us have to go out there and do something with our lives!
But what if there was a way to create a business that ran itself? Literally, ran itself? We don’t mean businesses like ATM or vending machine ownership either — those have been done to death, and while capable of generating “passive” income, still require someone to maintain them. No, this would be a business that would typically require a team of employees, each of which performs some particular function to reach a common goal.
You probably know where we’re going with this if you know anything about Expected X!
Agentic AI: Your New Startup Team
We recently came across a few articles on the internet diving into this topic. We thought it was worthwhile to not only analyze them, but add our own take since agentic AI is the backbone, if not philosophy, of Expected X’s take on the future of this field as a viable business investment. In essence, agentic AI might seem like the job-ending AI nightmare of so many professionals sleepless, anxiety-riddled nights. Not to worry though! While agentic AI systems can indeed be powerful compared to utilizing a single LLM, it isn’t going to eliminate your job….yet.
What Agentic AI Is
Agentic AI is both an LLM application paradigm and design framework:
In the paradigm sense, it’s the philosophy that LLM-based systems achieve better outcomes when they can retain a context-aware state, separate tasks among specialized agents, take actions via tools and API calls, orchestrate their process tasks, and collaborate on the final output of their given goal — wow, that’s a mouthful!
As a framework, there are several options available from LangChain’s LangGraph, Microsoft’s AutoGen, to crewAI. This is nowhere near an exhaustive list either as the interest in agentic AI has given birth to a variety of frameworks and SDKs.
While not exactly the same, traditional RAG systems (we say “traditional” because aRAG systems, agentic RAG, are, indeed, agentic) offered an early glimpse into a use case that extended the rather limited abilities of LLMs by themselves. Systems built with the abovementioned frameworks intend to extend the functionality of agentic AI towards the ontology of the paradigm…but let’s be clear, agentic AI is still a growing field.
What Agentic AI Isn’t
Although agentic AI frameworks are designed to aid developers build teams of autonomous “employees,” they still have a way to go before we’d feel comfortable letting them run complex businesses completely on their own. For everything that agentic AI aims to be (and we believe will be), currently there are limitations:
Agentic AI isn’t a complete replacement for humans and still requires human-in-the-loop feedback. One of the biggest challenges with using LLMs is that they will “hallucinate.” This can have a compounding effect in the context of agentic AI where one agent’s hallucination can impact the results it sends to other agents and the results those agents send along, and so on. Debugging agentic AI can be challenging as well (thankfully, all three frameworks we mentioned earlier — LangGraph, AutoGen, and crewAI — have built in debugging capabilities). So, those expecting a totally autonomous business of AI agents are rolling the dice.
Agentic AI suffers from the same lack of true “intelligence” as all LLMs — like those that underpin their functionality — do. Agentic AI, by itself, is still only capable of automating relatively simple tasks. Augmenting these systems with human-designed tools and capabilities can produce better outcomes, but relying on this type of custom tool creation and maintenance negates the autonomy we desire of not having to include humans in our daily business operations.
Agentic AI isn’t typically very cheap in a production environment. Most of the agentic frameworks default to OpenAI models which means API calls…and potentially a LOT of API calls depending on how your agents are structured. While this can be circumvented by pointing your agents towards OS models like Llama3, unfortunately OpenAI models still tend to produce some of the most optimal (on a side note, using different LLMs for each agent in your systems is typically a good idea to improve results).
Agentic AI isn’t typically quick to provide output. This isn’t necessarily a deal-breaker though as user expectations can be set ahead of time. Additionally, agentic systems are typically in use cases where time is not of the essence — they are designed to perform actions rather than provide immediate responses. For use cases that require immediate responses, fine-tuned models with “memory” are ideal, but this can also elicit exorbitant costs.
The Expected X Take…
We at Expected X believe that agentic AI is the future of AI application development. We’ve probably stated it more times than our audience cares to recall!. But let’s be clear…we are not calling it a panacea. Although the paradigm focuses on delineating tasks, processes, and tools among agents in a way that mimics how an organization would hire employees and build its org-chart, it is limited today. However, when you look at the advances in AI models occurring daily, it’s only a matter of time when you’re sipping Mai Tai’s in Hawai’i while your agentic AI business is roping in the revenue.
Ready to take your AI projects to the next level with agentic AI? Expected X is the partner you need—contact us today!
