A data platform manager we spoke with last month described their production database as ‘suddenly chatty’. The same number of users, the same applications, but query volume up 40% since they deployed an internal AI agent for customer support. Each agent request fans out into multiple database calls, retrieves from several sources, iterates, and repeats. The workload profile looks nothing like what their monitoring dashboards were built to track.
AI agents reason in loops, pulling data, checking results, pulling more data. A straightforward question like ‘which customers are at risk of churning’ might trigger a dozen queries across multiple tables before the agent returns an answer. Multiply that by hundreds of concurrent users and the database suddenly needs to handle traffic that used to be reserved for batch analytics jobs, except now it arrives throughout the working day with no predictable peaks.
The companies adapting fastest are hiring differently. Query optimisation skills that used to sit with specialist DBAs now appear in standard data engineering specifications. Experience with caching strategies and connection pooling appears alongside the usual pipeline requirements. One client rewrote a senior data engineer vacancy twice in six weeks, each revision adding more infrastructure depth after their existing team struggled to keep latency stable under the new load.
Candidates with database performance backgrounds are well positioned. Experience tuning PostgreSQL, managing read replicas, or building query-aware caching layers now commands attention it did not attract a year ago. Hiring managers would do well to plan infrastructure hires before an AI rollout reaches production, because retrofitting the team after agents go live means competing for the same scarce skills under time pressure.
Prompted by reporting from Datanami.