How to build agents with filesystems and bash
Sales call summarization costs dropped 75%. Here's how Vercel got Claude agents to navigate filesystems like code.
Why it matters
Vercel demonstrates a practical pattern for building cheaper, higher-quality AI agents by leveraging LLMs' native understanding of filesystem operations and bash—reducing infrastructure complexity while cutting costs 4x on real workloads.
The key facts
7 to knowSales call summarization agent cost reduced from $1.00 to $0.25 per call (75% cost reduction)
Deployed on Claude Opus 4.5
Output quality improved alongside cost reduction
Approach: filesystem tool + bash tool replacing custom tooling
Pattern applied to text-to-SQL agents
Agents leverage LLM training on code navigation for non-code structured data
Use case examples: customer support tickets, sales transcripts, CRM data, conversation history
Go to the source
Vercel Blogvercel.com
Publisher excerpt: Many of us have built complex tooling to feed our agents the right information. It's brittle because we're guessing what the model needs instead of letting it find what it needs. We've found a simpler approach. We replaced most of the custom tooling in our internal agents with a filesystem tool and…