TLDR: PostHog says it will train its own AI models using existing customer data in PostHog, with opt in by default for US cloud and opt out by default for EU. Training starts June 29, and users can opt out in org settings, but opted out accounts will miss AI features tied to that data.
Key Takeaways:
- PostHog is expanding AI features like AI-assisted replay analysis and the beta PostHog Code vision for proactive, self-driving product work.
- It will anonymize data and train models itself, not sell it or send it to third party model providers, with EU opt out and US opt in.
- Opting out blocks new AI features dependent on training data, so teams need to decide before training begins June 29.
This is a rare AI move that comes with a schedule and a switch. If you opt out, you are not just protecting data, you are also opting out of the smarter product promises tied to that data.
This is a rare AI move that comes with a schedule and a switch. If you opt out, you are not just protecting data, you are also opting out of the smarter product promises tied to that data.
Q&A
What happens to PostHog Code beta functionality for teams that opt out before June 29
PostHog says features dependent on training data will not be available to opted out users, so beta access will likely be limited for those accounts.
Why does PostHog choose opt in by default for US cloud but opt out by default for EU cloud
It says the split is meant to ensure sufficient training data for useful models while respecting obligations and user agreements tied to EU deployments.
How could training on session replay underlying data improve at scale versus replay analysis today
A model trained on the data powering replay detection could generalize across sessions, reducing per replay cost and letting teams diagnose issues more broadly.
What does synthetic user testing change about the timing of QA and release cycles
It shifts some testing earlier by flagging likely confusion and breakage from user behavior patterns, potentially reducing last minute review load for developers.
If PostHog only trains on existing data, what incentives does it have to keep improving model training methods
PostHog says the ideas are experimental and depends on iteration, so better training likely depends on finding which existing data signals actually produce useful predictions over time.
Top in AI
China expands certified AI chip list beyond Nvidia grip
MediaTek Dimensity 8550 pressures Googleâs Gemini Intelligence rollout
Google bundles Health Premium into AI Pro subscriptions
YouTube pushes AI labels into the viewerâs face
No comments yet. Be the first to share your thoughts!