d0
3 mentions across all digests
Vercel's internal text-to-SQL agent, simplified to use only bash command access, achieving 100% success rate while reducing latency and token usage.
Anyone can build agents, but it takes a platform to run them
Vercel argues production AI agents require dedicated infrastructure beyond models; the company showcases Sandboxes, Workflows, and AI Gateway for handling durability, isolation, and cost control at scale.
How to build agents with filesystems and bash
Vercel cuts agent-assisted sales call summarization costs by 4x (from $1 to $0.25 per call) using Claude Opus 4.5 with filesystem and bash operations instead of custom tooling, improving output quality.
We removed 80% of our agent’s tools
Vercel's d0 text-to-SQL agent jumped from 80% to 100% accuracy by removing 80% of its tools, proving modern LLMs thrive with minimal tooling rather than guardrailed complexity.