DevOps Paradox DOP 343: Your APIs Were Never Built to Be the Front Door
4 snips
Mar 25, 2026 Matt DeBergalis, CEO of Apollo GraphQL and longtime GraphQL practitioner, argues that AI agents are turning internal APIs into the customer-facing interface. He explains why typical REST APIs overload models and how GraphQL can orchestrate concise queries to reduce hallucination, cost, and latency. He also covers permission models, declarative infrastructure for many ephemeral APIs, and why docs must be written for models first.
AI Snips
Chapters
Transcript
Episode notes
APIs Become The New Front Door
- AI is turning internal APIs into the company's front door rather than websites being primary touchpoints.
- Matt DeBergalis explains models will call discrete capabilities directly, exposing APIs to end users in new ways.
Trim Responses Before Feeding Models
- Trim API responses and feed models only the exact fields they need to reduce cost, latency, and hallucinations.
- Use an adaptation/orchestration layer (AI orchestration) to transform verbose REST outputs into minimal model input.
REST Payloads Hurt Model Performance
- REST APIs return huge resource payloads that are fine for services but costly and noisy for models.
- Matt cites GitHub repo objects with hundreds of fields as harmful context that wastes tokens and causes hallucinations.