AI Automation / AI Agents
Build an agent that queries databases, generates insights, and creates visualizations automatically.
Why This Prompt Exists
Data analysis is bottlenecked by two things: writing SQL queries and interpreting results. An AI agent can do both — but needs the right tools and prompts.
You get:
- agents that generate wrong SQL (insecure or inefficient queries)
- agents that misinterpret results (correlation vs. causation mistakes)
- agents that can’t handle complex multi-table joins
- no visualization (just numbers, no charts)
- no way to ask follow-up questions (one-and-done analysis)
But data agents need structure:
- schema awareness: what tables and fields exist?
- query generation: natural language → SQL
- execution safety: read-only, timeouts, row limits
- insight extraction: what does this result mean?
- visualization: appropriate chart type for the data
Without design, data agents are dangerous.
This prompt designs safe, effective data analysis agents.
The Prompt
Assume the role of a data automation architect who designs AI data analysis agents. Your task is to create an agent that queries data and generates insights. Generate: 1. DATA SOURCES - Database(s) available - Key tables and fields (schema summary) - Update frequency (real-time / daily / weekly) 2. QUERY CAPABILITIES - Question types supported: [aggregation / filtering / trend / comparison / prediction] - Question types NOT supported: [e.g., "causation questions"] - Join complexity: [simple (2 tables) / moderate (3-5 tables) / complex] 3. SAFETY RULES - Read-only (no INSERT, UPDATE, DELETE) - Row limit (e.g., "never return more than 10,000 rows") - Timeout (e.g., "queries that run > 30 seconds are cancelled") - Sensitive data restrictions (e.g., "don't expose PII") 4. QUERY GENERATION PROTOCOL - Step 1: Parse natural language question - Step 2: Map to available tables/fields - Step 3: Generate SQL with comments - Step 4: Estimate result size (warn if too large) - Step 5: Execute after user approval 5. INSIGHT EXTRACTION - For each result: what does this mean for the business? - Statistical significance (if applicable) - Anomaly detection (what's unexpected?) - Comparison to historical baselines 6. VISUALIZATION RULES - Time series → line chart - Categories comparison → bar chart - Part-to-whole → pie/bar chart (prefer bar) - Distribution → histogram - Correlation → scatter plot 7. READY-TO-USE AGENT PROMPT - The system prompt for the data analysis agent INPUTS: Database schema (tables and fields): [PASTE OR DESCRIBE] Typical questions users ask: [E.G., "How many signups last week?", "What's our retention rate?"] User technical level: [NON-TECHNICAL / ANALYST / DATA SCIENTIST] Data volume: [SMALL (<1M rows) / MEDIUM (1M-100M) / LARGE (>100M)] RULES: - Always use read-only database connections (prevent accidents) - Set aggressive row limits for exploratory queries - Pre-validate SQL for syntax errors before execution - Flag results that exceed statistical or practical significance thresholds - Log all queries for audit and optimization - Provide explanations of results in business terms, not just statistics
How To Use It
- Always use read-only database connections for AI agents — one wrong update is catastrophic.
- Set aggressive row limits (1000 rows) for exploratory queries.
- Log all queries for audit and optimization.
- Provide business explanations for results, not just statistics.
- Flag results that might be statistically significant but practically meaningless.
Example Input
Database schema:
“Users table (user_id, signup_date, plan_type, country). Payments table (payment_id, user_id, amount, date).”
Typical questions users ask:
“Monthly revenue, signups by country, retention by plan type”
User technical level:
“NON-TECHNICAL”
Data volume:
“MEDIUM”
Why It Works
Most data agents try to answer any question — which leads to wrong SQL, misinterpreted results, and dangerous queries.
This framework improves outcomes by forcing:
- schema awareness (what data is available?)
- query capability boundaries (what questions can it answer?)
- safety rules (read-only, row limits, timeouts)
- insight extraction (what does the result mean?)
- visualization rules (right chart for the data)
Great data analysis agents don’t pretend to answer everything — they answer a defined set of questions safely and clearly.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.