Product analytics tilts towards agents and SDK sprawl

On 22 Oct, Amplitude said Vijaye Raji, founder of Statsig, had joined its board, with press coverage also noting his role as CTO of OpenAI. In a category built on measuring user behaviour, that appointment reads as a strategic signal around experimentation, product analytics and AI systems converging.

The product surface followed the same line. From 21 Oct to 24 Oct, Amplitude added pages for agent analytics across frameworks including Django and FastAPI, AI agent evaluation metrics, AI agent evaluation comparisons, and a Java agent for analytics and feature flags. Its Amplitude-AI-Node releases on 20 Oct and 25 Oct improved TypeScript ergonomics, JSON serialisation, numeric preservation and redaction of sensitive information.

PostHog was just as direct. On 15 Oct it announced five new MCP server tools, including Signals, Replay vision and Annotations. On 23 Oct it introduced AI observability for foreground Gemini interaction calls, streamed responses and tool activity, plus WebMCP tool-call analytics. Heap also added a blog page called Introducing Heap MCP on 23 Oct, while Mixpanel created an ai-plugins repository the same day. The category is not merely adding AI labels. It is building instrumentation for AI workflows.

The fortnight was unusually dense with runtime and SDK activity. Amplitude created or surfaced repositories including Amplitude-Flutter on 21 Oct, Amplitude-Python and experiment-jvm-server on 15 Oct, and AmplitudeUnified-Swift and AmplitudeCore-Swift on 18 Oct. Its site additions also moved through WordPress pages on 17 Oct, Astro and Framer analytics pages on 20 Oct, Flutter and Go pages on 23 Oct, and Java agent material on 24 Oct.

PostHog spread even wider. It created or introduced repositories for posthog-android, posthog-go, posthog-js, posthog-ios, posthog-python, posthog-rs, posthog-ruby, posthog-elixir, posthog-unity, posthog-roblox and posthog-php across the period. It also shipped posthog-js updates on 20 Oct, 22 Oct and 23 Oct, alongside repeated @posthog/ai releases.

Mixpanel joined the same developer distribution contest with mixpanel-android on 17 Oct, pi-subagents on 19 Oct, mixpanel-headless v0.4.0 on 21 Oct, ai-plugins on 23 Oct and mixpanel-unity on 24 Oct. Statsig added statsig-ai-node on 17 Oct, android-sdk on 20 Oct, dart-sdk on 22 Oct, js-client-monorepo and statsig-kit on 23 Oct, and sendgrid-webhook on 25 Oct. The message is plain: product analytics is being pulled closer to application code, not left inside the analyst workspace.

The sequencing mattered more than any single page. Amplitude's site did not pivot once. It marched: React Router pages on 15 Oct, WordPress pages on 17 Oct, 28 product pages around Astro and Framer on 20 Oct, agent analytics pages on 21 Oct, Django, FastAPI, Flutter and AI evaluation pages on 22 Oct, Flutter and Go pages on 23 Oct, and event plus Java agent pages on 24 Oct and 26 Oct. At the same time, it removed pages tied to venture and portfolio programmes, Japanese product pages, WordPress tracking metrics, Wix tracking metrics and a Daily Active Users page related to Wix.

PostHog's content motion was even heavier. On 15 Oct it added 75 product pages and removed 68 documentation pages. On 17 Oct it added 21 documentation pages and 12 product pages while removing 34 documentation pages. On 21 Oct it added one blog page, 31 documentation pages and 32 product pages, while removing 12 documentation pages and 52 product pages. On 24 Oct it added documentation for AI evaluations, data modelling, error tracking and web analytics, while removing product and pricing pages related to error tracking and experiments.

That is the quiet quarter of this market: the first proof is often not a press release, but a run of sitemap edits, repository creation and documentation swaps. Operators who wait for the launch post see the positioning after it has already been assembled.

On 25 Oct, Mixpanel reported degraded service for requests to decide.mixpanel.com. The incident lasted around 453 minutes and moved to postmortem status after resolution. For a service involved in decisioning, even a resolved incident becomes part of how buyers evaluate operational maturity.

Amplitude had several resolved incidents in the period. On 18 Oct, ingestion systems lagged for around 116 minutes. On 21 Oct, Export API 500 errors affected multiple organisations for about 8 minutes. On 23 Oct, Streaming to External Destinations Unavailable lasted around 278 minutes, and a website outage lasted about 1 minute. On 25 Oct, delayed or blocked event delivery to Amazon Kinesis lasted around 68 minutes. The reported impact varied, including minor and none, but the pattern keeps pipeline resilience in the foreground.

Package demand signals were also noisy. On 18 Oct, reported weekly downloads fell sharply for several Amplitude, PostHog, Statsig, Heap and Mixpanel packages. By 24 Oct, Statsig's statsig-node package had risen from 322,403 to 451,616 weekly downloads, and Heap's React Native autocapture packages rose 121 percent and 108 percent. One week of package movement is not a thesis, but it is a useful reminder that adoption signals in this category can move faster than positioning copy.