Agents aren’t all you need
We founded Parcha in early 2023, when "AI agents" built with language models emerged. We were inspired by demos showcasing how LLMs and code could dynamically create and execute complex plans using agentic behavior. Frameworks like AutoGPT and BabyAGI emerged, and people started applying this technology to numerous use cases. The demos looked promising, and the excitement around autonomous AI agents was palpable. Soon after, we saw startups emerging for horizontal (AI Agent platforms) and vertical (AI Agents for specific industries) use cases like Parcha..
We founded Parcha with the thesis that AI agents could automate many operational processes in fintech and banks. Our initial approach was to take a standard operating procedure (SOP) that a human follows to perform an operational process and have an AI agent read and execute those steps autonomously. We developed an agent that would read the SOP and assume the role of the operations expert performing this process today, such as a compliance analyst.
The Parcha agent would generate dynamic plans based on the SOPs provided by our design partners. These plans would guide the agent through each process step, utilizing various tools and commands to execute tasks and record outcomes in the scratchpad. Once it has performed all the steps, the agent will use the SOP and the scratchpad to make a final determination, such as approving or denying a customer, and deliver the result to the end customer. We aimed to create a flexible and adaptive system capable of handling a wide range of workflows with minimal human intervention.
