EXPOSED
Most of the day is schema design, SQL/DDL authoring, query tuning, ETL mapping and documentation — all things LLMs already produce at usable quality for standard patterns, which compresses the routine tier of this job hard. What persists is ownership of consequential, ambiguous calls: data model tradeoffs that lock in a decade of application behavior, partitioning and capacity strategy under real load, migration cutovers where a mistake corrupts the business record of truth, and compliance posture for regulated data. There is no license and no signature requirement, so the moat is accountability and organizational trust rather than regulation.
Mixed — a routine tier and a judgment tier. Normalization to 3NF, index selection, writing DDL and stored procedures, dimensional modeling for a warehouse, and drafting data dictionaries are all pattern work a model reproduces from a requirements paragraph — what sits above 6 is the un-writable part: reverse-engineering an undocumented legacy schema where column names lie, negotiating with three application teams over who owns the customer record, and sequencing a zero-downtime migration against a live OLTP system.
Fully desk- and screen-based. The 2 is for the rare rack-and-console day — standing up a cluster in a colo, sizing physical storage, or being in the datacenter during a cutover window — because everything else is DBeaver, dbt, Terraform and a Zoom call.
No licence, no signature requirement. No state licenses database architects and no statute requires a named human to sign off a schema change; the 4 reflects only the practical gatekeeping of vendor certifications (Oracle OCP, AWS/Azure data credentials) and SOX/HIPAA change-control records that put your name on the approval ticket without putting you personally on the hook.
Some relationship component. Business stakeholders rarely know who designed the schema they query, but the 8 comes from the internal standing you accumulate — being the person the application leads call before they add a table, and whose word on a retention or PII decision the compliance team accepts without re-litigating it.
Exists to be accountable for ambiguous calls. A 15 is warranted because the calls you own are irreversible at business scale: choosing a sharding key or surrogate-key strategy that application code will assume for a decade, deciding whether a migration cutover proceeds or rolls back at 3am with the record of truth in flight, and setting encryption, masking and retention posture for regulated data where the wrong choice becomes a breach report rather than a bug ticket.
Has AI actually changed your work?