· 6 min read
agentic-ai
ClickHouse's Agentic SQL Benchmark Shows the Bottleneck Moved Upstream of SQL
ClickHouse benchmarked 28 models on 201 real analytics questions, letting agents take up to 60 turns each. SQL execution errors are nearly gone. The dominant failure, 53-82% of the time, is the agent building the wrong plan before it writes a single query.
Read more