AI SRE needs better observability, not bigger models (clickhouse.com)

Recent insights reveal that the effectiveness of AI-driven Site Reliability Engineering (SRE) tools is hampered not by the intelligence of the models, but by the inadequacy of their underlying observability infrastructures. Many AI SRE solutions, which aim to automate incident response using large language models (LLMs), fall short due to limitations in data retention, high cardinality, and slow query speeds. This issues lead to a failure in accurately diagnosing incidents, as the models struggle with incomplete data and lack the context necessary for reliable root cause analysis. A proposed solution emphasizes the use of ClickHouse, a columnar database ideal for high-cardinality analytics and long-term data retention, as the backbone for AI SRE tools. ClickHouse addresses the critical challenges of traditional observability systems by enabling longer data retention, supporting billions of unique values without compromising performance, and ensuring sub-second query speeds. The core philosophy encourages leveraging AI to assist human engineers by enhancing their contextual understanding during incidents, rather than relying solely on automated remediation, ultimately improving incident resolution times and promoting a more informed human-in-the-loop approach.