Sep 18, 2023

Since ChatGPT came out in November 2022, Generative AI has been in peak hype cycle.  Just about every business is trying to figure out how to use this new technology.  Most of the focus is on LLMs (or “Large Language Models”) and how to fine tune models and slap an API on them for domain-specific applications.  While undeniably critical, we believe that the majority are overlooking the true source of impending revolutionary value - AI Agents empowered by LLMs.

We decided to break down some key differences and the evolution from Software to LLMs and ultimately to AI Agents.


Pre-GenAI Software: Where We Were

Before the advent of Generative AI, traditional software predominantly relied on clean, structured data, faltering when faced with ambiguous or unstructured inputs. These programs are bound by deterministic processes, consistently producing the same output for a given input based on precise, rule-based algorithms. Their operations are set within rigid parameters, with their flexibility only extending as far as the original programmer's anticipation. Critically, these systems lack the capacity for human-like reasoning; decisions are hard-coded, devoid of the nuanced judgment and adaptability characteristic of human thought.



Large Language Models (LLMs): Where We Are Today

We are in the age of LLMs, like OpenAI's GPT series. These models bring a significant paradigm shift where they can broadly take unstructured inputs as natural language and generate output (text, code, pictures, etc.) so dynamically that it seems to reflect human reasoning.  


LLMs, while amazing, are still only passive tools.  They do not proactively take action, they simply respond to an input.  It’s for this reason that LLMs alone will not bring on the mass automation that everyone is anticipating with Generative AI


Enter AI Agents…


AI Agents (Autonomous Software): Where We Are Going

AI Agents stand at the forefront of modern software evolution. These sophisticated systems are distinguished by their ability to tap into the capabilities of LLMs for intricate reasoning processes. Leveraging LLMs, AI agents can manage and interpret unstructured data and complete non-deterministic paths - i.e. “Autonomous Software”.  


Agents are not just static applications that follow predetermined rules. Instead, they continuously learn, self-enhance, and nimbly adapt to shifting circumstances and environments.  


Example - Sales Outreach

Sales outreach is the perfect example to show the differences between existing software, using LLMs, and AI Agents.   Let’s take the CRM Hubspot, for example.  Specifically, let's look at their “Sequences'' tool where you can automatically send pre-written emails and follow up emails to prospective customers.  


Existing Software: In order to start this process, you first have to obtain emails/contact info, next you have to write the emails, set exactly which people you want to send the email to, determine how many days apart to send each follow up email, and decide on a few more settings.  Hubspot does offer automation tools to automatically enroll contacts into sequences, but these are again based on pre-set conditions (e.g. if a contact is considered a lead, enroll them into the sequence.


LLMs:  Today, the way most applications are implementing LLMs is by putting an API on top of GPT and allowing users to use the LLM like ChatGPT but in their normal workflow.  So, in our sequencing example, the user could quickly generate email copy or an image based on a couple of bullets.  While helpful, this is not the holy grail promise of GenAI.


AI Agents:  With just a few simple prompts, and agentic system can autonomously obtain contact information (via lead database like ZoomInfo or crawling the web), organize it in your CRM, create email copy, setup Sequences, A/B test, track effectiveness (open rate, response rate, etc.), iterate, and so on.  


Taking this a step further, our vision is that each department of a company will have teams of AI Agents.  These Agents will coordinate to carry out complex tasks leading to common goals.  So let’s say a new product is ready for launch, the product team’s agents will coordinate with the marketing team’s agents for a GTM strategy, the marketing Agents will coordinate with the Sales team’s Agents who will take the steps described above.


Conclusion

While the recent advancements in the realm of Generative AI, have been nothing short of impressive, it's the AI Agents that hold the key to the promised efficiency gains. Their integration with LLMs is poised to automate a myriad of tasks, both complex and mundane, which have traditionally consumed vast amounts of time and resources.