Managing Memory for AI Agents

Chapter 4. Navigating Agent Trade-Offs: Custom Builds, Frameworks, and Hosted Solutions

Digital organizations shift the primary focus of integration from the systems to the capabilities, emphasizing clean interfaces over those capabilities.

Brandon Byars

Whether you are an insurance company or a startup, the fundamental considerations at the start of the process of deploying an AI agent are largely the same: build versus buy, frameworks versus customization, and of course, cost and scalability concerns. If these trade-offs and considerations sound familiar, they should. Despite AI agents being a wholly new category of technology, their fundamentals are those of most software products. That should give us heart because we have decades of precedent upon which we can rely to build resilient solutions.

The Developer’s Dilemma: Framework Versus Custom Architecture

Not all projects involving or centered around AI agents require the same tooling considerations. Ultimately, the choice between, say, using LangGraph to build out an agent and turning to a hosted service such as Glean AI comes down to one thing: scope. Is the project you’re about to undertake existential to your company’s future, or is it experimental and exploratory? How much data will you need to ingest and process regularly? Will your initial user base begin as large or small? Is this purely an internal tool, or will it be customer facing? Does your company already have a robust infrastructure supporting AI agents, ...