Imprint is developing internal agent workflows to enhance their capabilities, specifically for co-branded credit card programs. The author shares insights from their experiences, breaking down several key areas of focus. These include skill support, context window management, and iterative prompt refinement. The aim is to document practical problems and solutions encountered while evolving these workflows, rather than present an expert guide. To learn about building agents, the author suggests a hands-on approach. Start by reading foundational materials like Chip Huyen's "AI Engineering" to grasp how Large Language Models (LLMs) function. Then, create a simple script using an LLM API, gradually extending its capabilities with tools for file searching and context management. Implementing features like a virtual file system and post-workflow evaluations will solidify understanding. While existing frameworks like OpenAI's and Claude’s may offer some benefits, the author believes building your own from scratch provides deeper insights. The author emphasizes that every company, even those not directly involved in AI, should explore internal agent development. This work can lead to significant improvements and is a cost-effective way to stay ahead of potential changes in AI technology. Having a couple of engineers focused on this area can help mitigate risks associated with rapid advancements in AI.