AI Agents6 min readUpdated 2026-06-29
Wise + n8n AI Prompt Library
Battle-tested AI prompts for classifying, drafting, and enriching Wise records inside n8n.
Key takeaways
- Prompts are code.
- Structured outputs always.
- Cache aggressively.
- Version like software.
Prompts are code. Here's the versioned library we deploy for Wise + n8n across classify, draft, enrich and summarize tasks.
Classify prompt
Structured-output schema, temperature 0, one-shot example, deterministic. Costs cents per 1k records.
Draft prompt
Role + brand voice + explicit constraints. Human-in-loop approval required.
Enrich prompt
Extract JSON matching the Wise field schema. Reject anything that doesn't validate.
Summarize prompt
Bullet output, max length, focus areas explicit. Cache identical inputs.
Frequently asked questions
- Where do I store prompts?
- Postgres or Notion with version numbers.
- Testing?
- Golden dataset of 20-50 examples; diff on every prompt change.
- Models?
- GPT-4o-mini for scale, GPT-4o for quality-sensitive drafts.
- Guardrails?
- Schema validation + human approval for writes.