Product analytics turns into agent infrastructure

Mixpanel made the cleanest move. On 5 Oct it announced Agent Intelligence, with coverage highlighting outcome analytics for AI agent interactions. The launch had already been trailed in the product surface: on 1 Oct Mixpanel added an agent intelligence documentation page, and on 26 Sep it introduced seven documentation pages covering AI agents and data warehouse connectors.

Amplitude is moving in the same direction, but through a broader content and M&A story. On 28 Sep it added a State of Agentic Analytics product page. On 2 Oct it added a blog page titled What AI Engineers Do. Dealroom reports on 1 Oct also listed Amplitude’s acquisitions of Command AI for $33.6 million and Kraftful for $3.8 million.

PostHog framed the shift differently. On 28 Sep its homepage hero changed from Make your product self-driving to Your product's context layer, while the subtext moved from a specific demo message to a broader claim about ingesting and storing data for querying and action. On 1 Oct it added a blog post titled How AI Agents Behave, and its @posthog/ai package moved from 8.13.3 to 8.13.4 on 30 Sep.

The point is not that every product analytics vendor now says AI. The observed pattern is more specific: agent behaviour is becoming a measured workflow, not a marketing page. The winners are trying to own the loop from user action, to agent action, to outcome.

PostHog’s documentation churn was the loudest operational signal of the fortnight. On 28 Sep it added 44 CDP source pages and removed 45 documentation pages. On 1 Oct it added 44 documentation pages and removed 42 documentation pages. On 2 Oct it added 34 documentation pages and 26 product pages, while removing 60 documentation pages. By 3 Oct and 5 Oct, the pattern had narrowed towards AWS sources for API, CDP and data warehouse use cases.

That is not normal brochure maintenance. It reads like a product surface being re-indexed around the data layer. The same direction showed up in the product itself: on 28 Sep PostHog introduced BigQuery service account imports, allowing data warehouse imports through a shared Google Cloud service account rather than an uploaded JSON key.

Mixpanel is also stretching into the integration layer. On 30 Sep it added documentation for cohort sync with Wingify, no-code experiments, and multiple MCP connections including GitHub and Jira. On 26 Sep it added data warehouse connector documentation alongside troubleshooting pages for report building, data export, identity management and user data protection.

The useful signal was the sequence. PostHog’s homepage repositioning landed on 28 Sep, the same day as its BigQuery import update, before several days of documentation replacement. Mixpanel’s agent documentation appeared on 1 Oct, after the MCP and warehouse connector material on 30 Sep. Product analytics is being rewritten as plumbing, then described as intelligence.

The GitHub and package activity was unusually dense. Mixpanel released version 0.3.0 of mixpanel-headless on 2 Oct, adding built-in API help and mobile session replays. The same event reported top-repository stars rising from 675 to 2,568, and new mixpanel-android and mixpanel-js repositories. On 1 Oct, Mixpanel also released version 1.6.0 of its iOS session replay package, adding a beta wireframes feature and a fix to prevent event processing during exponential backoff.

PostHog kept shipping across clients and agent surfaces. On 27 Sep it released posthog-js 1.434.15, @posthog/ai 8.13.2, @posthog/cli 0.18.7 and @posthog/react 1.11.2. On 29 Sep it released posthog-js 1.434.17 and @posthog/cli 0.18.9. On 2 Oct it shipped posthog-js 1.435.6 and @posthog/ai 8.13.4. Across the same period, it created repositories including posthog-ios, posthog-android, posthog-flutter, posthog-kmp, sdk-specs, hoglake, trino and duckgres.

Statsig also widened its developer footprint. On 28 Sep, registry activity showed @statsig/js-client at 4,052,316 weekly downloads after a 63% rise, statsig-node at 901,922 after a 71% rise, and statsig-node-vercel at 166,105 after a 73% rise. On 1 Oct it launched eight repositories including statsig-go-core, statsig-php-core and statsig-ai-node.

Amplitude’s developer work was steady rather than singular. On 25 Sep it added repositories including rrweb, Amplitude-Java, Amplitude-Swift and Amplitude-MCP-Analytics-Node. On 30 Sep it released @amplitude/analytics-browser 2.47.2 and @amplitude/analytics-node 1.5.76. On 1 Oct it released Amplitude-Kotlin 1.33.0 with streaming analytics available as a separate Maven artefact.

Mixpanel announced a $200 million Series C led by Bain Capital Tech Opportunities on 28 Sep and 29 Sep. Its hiring signal moved in parallel: on 30 Sep it had four new postings across engineering and customer-facing roles, and on 1 Oct it had three new postings including Renewals Manager and Strategic Finance Associate in Singapore, New York City and San Francisco.

Amplitude’s commercial week was more mixed. On 25 Sep it removed the first 2 million events per month free from the Plus tier. On 25 Sep it also reported three data delivery incidents affecting Export Destinations and Snowflake Data Share, with observed durations of about 94 minutes, 365 minutes and 364 minutes. On 29 Sep it reported a resolved minor data export incident lasting around 324 minutes.

Heap’s fortnight was defined by integration and security notices rather than positioning. On 24 Sep it reported two resolved incidents: a Heap Connect Syncs issue lasting about 160 minutes, and third-party integration event latency lasting around 952 minutes. On 3 Oct its status page reported a major incident in which data was not syncing for the Salesforce integration, with the issue identified and no duration provided. On 30 Sep, two CVEs originally published on 4 Sep were observed naming Heap, with CVSS scores of 7.8 and 7.6.

Pricing also tightened at Mixpanel. On 1 Oct, Root cause analysis and Metric Trees were removed from the Enterprise tier. Taken with Amplitude’s 25 Sep Plus change, the category’s packaging is becoming less generous just as vendors ask buyers to believe that analytics now includes agents, replay, experimentation, warehouse sync and outcome measurement.