Data4 min read
Data quality is infrastructure, not a chore
Why cleaning, validating and normalizing data belongs behind an API instead of being rewritten in every project.
Almost every application contains a quiet layer of data-handling code: trimming strings, checking formats, catching duplicates, reshaping records into whatever the next system expects.
It is rarely designed. It accumulates, usually in response to a bug.
The cost of rewriting it per project
Each reimplementation has slightly different rules, which means the same organisation ends up with several definitions of a valid record.
- Rules drift between services.
- Fixes land in one codebase and not the others.
- Nothing is testable as a single thing.
Agents raise the stakes
An AI agent acting on inconsistent input does not hesitate — it acts confidently on bad data. Treating cleaning, validation, normalization and deduplication as shared infrastructure gives applications and agents the same guarantees.