Product analytics, observability, and AI development tools
PostHog
PostHog builds tools for product engineers to understand customer behavior, investigate problems, and improve software. Its platform combines product analytics, session replay, feature flags, experiments, error tracking, and AI observability. A context warehouse brings product data together with external business data so people and agents can query it directly. Its self-driving development approach connects that context to automated problem diagnosis, bug fixes, and pull requests, with access through web, Slack, desktop, CLI, and MCP interfaces.
Founded in 2020 by James Hawkins and Tim Glaser, who serve as co-CEOs, PostHog began as open-source product analytics for developers. Its engineering approach includes HogQL, a translation layer over ClickHouse SQL that simplifies access to event and person properties and automatically adds joins across tables. Engineers can use it to filter events, extend analytics, and write custom queries within the same platform.
As of August 2026, the company reported more than 500,000 teams, with 98% of customers using PostHog free. Paid products use usage-based pricing with monthly free tiers. PostHog became cashflow positive in December 2024 and announced a $75 million Series E in 2025, led by Peak XV Partners at a $1.4 billion valuation.
2 talks
Newest first
2 speakers at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here
- LLM codegen fails and how to stop 'em
Start here to understand why stale model knowledge causes integration failures and how prose instructions can retain value as models improve.
Danilo CamposAI Engineer Europe 2026
- Self Driving Products: Product Signals to Pull Requests
Follow this talk for the execution setup: the Claude Agent SDK, Modal sandboxes, and an MCP server supplying additional context during automated development.
Joshua SnyderAI Engineer Europe 2026
Messages from the stage
Guide implementation with concrete references
Campos describes current documentation, lightweight representative projects called Model Airplanes, and controlled implementation steps as ways to reduce hallucinated APIs and architectural mistakes. His discussion balances useful agent boundaries against excessive constraint.
Investigate signals before implementing fixes
Snyder's pipeline screens incoming signals for unsafe instructions, normalizes and semantically groups them, then assigns investigation to a research agent. Evaluation against representative production data is part of assessing the system.
Affiliations reflect each recorded session, not necessarily current employment.