AI for Database Management

TL;DR

• AI can generate SQL, suggest schemas, and draft migrations.
• It will write syntactically correct queries that perform terribly. Always verify execution plans.
• Use AI for drafts. You own correctness, indexes, and production safety.

Databases are unforgiving. A bad query can lock tables. A bad migration can lose data. AI accelerates the drafting — you ensure it's safe.

Tools (2026): There's no dominant "database AI" product. Most teams use general agents: Claude Code (~$20/mo Pro, 200k context) and Cursor ($20 Pro / $200 Ultra) for SQL, migrations, and schema design within full-stack workflows. Retool adds AI-assisted queries for internal tools. n8n chains LLMs with databases, Slack, and HubSpot for workflow automation. Stick with Cursor, Claude Code, or Copilot for DB work—they fit your existing stack.

Query Optimization

Good use cases:

• "Optimize this query. Database: Postgres 15. Table sizes: [approx]"
• "Suggest indexes for this query pattern"
• "Explain this execution plan. What's the bottleneck?"

Critical checks:

• Execution plan — Run EXPLAIN. AI may suggest an index that doesn't help or makes things worse.
• Cardinality — AI doesn't know your data distribution. Low-cardinality columns, skewed data — verify.
• Locking — AI may suggest changes that increase lock contention. Consider MVCC and isolation.

Workflow: Generate → EXPLAIN → test on staging with prod-like data → then prod.

Schema Design

Good use cases:

• "Design a schema for [domain]. Requirements: [list]"
• "Suggest normalizations for this denormalized structure"
• "Add audit columns to this schema. Pattern: created_at, updated_at, etc."

Cautions:

• AI may over-normalize or under-normalize. You know your access patterns.
• Naming and conventions — align with your org. AI will use generic names.
• Future evolution — migrations, backfills — AI may not consider. You plan.

Migrations

Good use cases:

• "Generate a migration to add column X to table Y. Zero-downtime preferred"
• "Draft a migration from schema A to schema B"
• "Suggest a backfill strategy for this migration"

Critical:

• Data safety — Never run AI-generated migrations on prod without review. Test on copy of prod.
• Rollback — Does the migration have a rollback? AI may not include it.
• Concurrency — Long-running migrations on large tables need special handling. AI may not account for it.

Documentation

Good use cases:

• "Document this schema for a data dictionary"
• "Create a README for this database"
• "Explain this stored procedure"

Verify against actual objects. AI can hallucinate columns or behavior.

-- AI suggested "optimization" — always verify with EXPLAIN
-- Before: Full table scan
EXPLAIN ANALYZE
SELECT * FROM orders WHERE status = 'pending';

-- AI might suggest an index on status
-- Run EXPLAIN before AND after adding the index
-- Verify: Does it use the index? Did timing improve?
-- Check: What about your data distribution? Low cardinality = index might not help.

Do This Next

1. Feed AI a slow query you've seen. Get optimization suggestions in Claude Code or Cursor. Run EXPLAIN before/after. Verify the improvement.
2. Have AI draft one migration (even a simple one). Review for safety, rollback, and concurrency. Run in dev only.
3. If you use Retool or n8n: Connect an LLM to a read-only DB for internal query assistance; never expose write access in prompts.