A LinkedIn post from PostHog highlights internal usage data from its MCP analytics platform, offering insight into how AI models and tools are being adopted by agents. The post suggests that Anthropic’s Claude family, particularly Opus 5.5, currently accounts for a majority of observed model usage, outpacing combined OpenAI models over the same period.
According to the post, Opus 5.5 has rapidly become the dominant model since its September launch, while OpenAI usage is fragmented across several versions. The company’s LinkedIn post also notes that the gap is narrower in code-focused workloads, where Claude Code and OpenAI’s Codex are more evenly matched, indicating competitive dynamics in developer-centric AI applications.
The post indicates that SQL is the most frequently used tool by agents, with the majority of sessions executing queries and SQL representing over half of overall tool usage. This pattern suggests that data-access and analytics workflows remain central in agent deployments, which may inform where PostHog and other vendors prioritize product development and integrations.
PostHog’s LinkedIn content also points to a higher call volume and error rate in agent-driven requests, though the company characterizes the error rate as acceptable given experimental use cases. For investors, these observations may imply that agent-based automation is still in a maturation phase, potentially requiring continued investment in reliability, observability, and tooling.
As shared in the post, agents frequently provide reasons for their actions and metadata about tool usage, enabling more granular analytics on behavior and performance. This focus on instrumentation and feedback loops could strengthen PostHog’s positioning in AI-native product analytics, supporting differentiation versus traditional web analytics platforms and potentially enhancing its long-term competitive profile.