How LLMs deployed as AI Agents are going to transform knowledge work

In the past year, large language models (LLMs) such as OpenAI’s GPT series and others have improved significantly, surpassing human performance on many fundamental human tasks. 

This milestone is noteworthy, and has enormous implications for the future of work, because LLMs can be used for tasks that require intelligent reasoning — an ability that, throughout all of history, has required a human.

There have been a few enablers for this change: 

1. Models trained on vast quantities of data: The performance of LLMs improved as the size of the model and amount of training data increased, so without improving algorithms we can still achieve better performance.

2. Transformer architecture: This deep learning approach allows the model to weigh the importance of different elements in an input sequence when producing an output for each element, resulting in improved performance.

3. Increasing context windows: A context window is the amount of information you can pass into the prompt. The larger it is, the more direction you can provide to an LLM, and the better the outputs for more complex tasks. 

In addition, LLMs gained more mainstream popularity once they were taken out of academic environments and introduced to everyday consumers, such as with ChatGPT.

These advances in AI and LLMs not only enable us to create software differently but also make it possible to build products unlike anything we’ve ever imagined before — and have led to a new form factor for software capable of automating tremendous amounts of manual work: AI Agents. 

Agents will reshape traditional knowledge work in ways we could only imagine a few years ago.

What is an AI Agent, and what is it good at? 

AI Agents are LLMs tasked with achieving objectives. They extend the capabilities of LLMs by having access to tools such as APIs, databases, web searches, and even other sub-agents. They also have the ability to iterate and improve autonomously through reflective loops, learning from outcomes and refining their approach over time. Agents will increasingly coordinate and collaborate with other agents, achieving levels of optimization and opportunity that are difficult to comprehend the ultimate impact of.

Their ability to interact with and learn from their environment makes them well-suited to a variety of workflows. They excel at handling tasks that involve a degree of standardization but variable enough to require humans.

Agents will transform many job functions, but we believe risk teams at financial services companies in particular stand to benefit greatly from Agents. The role of a risk team is inherently a data problem, with a lot of repetition, requiring coordination between multiple tools, identifying trends, and communicating analysis. Agents, and essentially software systems, are now well suited to the parts humans were previously much better at. They’re also more reliable, scalable, and faster.

For example, within a Risk team it is very common to have a team of Fraud Analysts who review escalated applications that triggered a fraud signal. The analysts would then review the customer-provided information and data returned from vendors in their fraud stack, complement it with web searches, following an internal checklist, then synthesizing the case, making a decision, and communicating it to other colleagues. An Agent can now carry out all of the steps up to the making of the final decision.

The future of knowledge work

The deployment of Agents in knowledge work will lead to large shifts in productivity and creativity. As they take on repetitive and mundane tasks, humans can focus on higher-order, creative activities. Knowledge workers will still need to be involved in final decisions in critical and regulated use cases. Agents are an opportunity for humans and machines to work together better than ever before, amplifying each other's strengths.

Companies will become more productive, leading to increased output and, consequently, an increase in jobs and economic growth. The pace of progress will continue to surprise us, and risk teams that adopt Agents will pull ahead of those that don't. At Inscribe, we want to lead the way in deploying AI Agents in financial services. If you’re a risk leader or financial services operator interested in learning more, book time with us here.