PostHog · AI product · Senior · 45 to 60 minutes

How should PostHog bet on AI agents?

PostHog is a developer focused product platform with tools like analytics, feature flags, and session replay that help software teams understand and improve their products. Answer this product management case and get it graded line by line against a weighted rubric.

Product strategy
AI judgment
Prioritization
Systems thinking
Executive communication

The situation

The CEO has asked you, a senior PM working across the product line, for a recommendation on PostHog's AI direction. Customers and prospects are increasingly asking how AI will help them get more value from their product data and workflows, and the market is filling up with copilots, chat interfaces, and early agent claims. Inside PostHog, there are several plausible paths: add more AI answers inside existing tools, make the platform easier for software agents to use, build experiences that proactively surface work instead of waiting for a prompt, or hold back until the technology is more reliable. Assume leadership wants a recommendation for the next 12 to 18 months. The decision is hard because LLMs still make mistakes, autonomous behavior can break trust, and a company level bet could reshape APIs, permissions, onboarding, and how each product works. At the same time, doing too little could leave PostHog reacting to the market instead of shaping it.

What to produce

A 1 to 2 page strategy memo.

Your answer must

Define the main customer problem PostHog should solve with AI first, and why it matters now.
Recommend one primary strategic direction, plus what PostHog should not do yet.
Lay out a phased plan with near term foundations, first customer facing step, and later expansion.
Explain how your plan handles LLM limits, trust, permissions, and the choice between reactive help and proactive action.
Propose success metrics and decision checkpoints that would tell leadership to double down, adjust, or stop.

Constraints

Work from PostHog's current multi product platform and existing customer base.
Do not assume LLM reliability improves enough soon for fully unattended important actions.
Any recommendation must preserve clear user control, permissions, and auditability when software acts on a user's behalf.
You can recommend sequencing, but not several large AI bets launched at once.

Format

Use headings: recommendation, why now, phased plan, risks, metrics.

How PostHog actually played it

The PostHog CEO lays out why the company is shifting toward agents: making the product usable by agents, moving from reactive to proactive, and building long running agents. The post is honest about the limits of LLMs and the pull of fear of missing out. Good for seeing how a mature product company sets an AI direction.

Separate real progress from fear of missing out when choosing an AI direction
Make the product agent accessible before adding new agent features
Move from answering questions to proactively surfacing work

Built from PostHog's next chapter by James Hawkins.