04.7.1 · Skill in AI-native engineering
LLM integration
Calling language models from a product reliably.
Integrating an LLM means prompts, model choice, streaming, retries, timeouts and cost control. Prompts are code and deserve versioning and tests. The model is a dependency that changes, so the integration has to absorb that.
Resources
2- docsStart herePrompt engineering overview ↗Anthropic docs · 5 min
Where to start when a model's output is not good enough, before changing models or adding complexity.
- articleStart hereBuilding effective AI agents ↗Anthropic · 2024 · 12 min
When a simple workflow beats an agent, and the handful of patterns that cover most real systems.
Tools
none yetTools for this topic are being researched.