Learn AI Observability with your agent

Learn with examples from your own product and data.

Start with a learning path for your product, or choose a topic below. Each prompt uses your code and data to teach one concept at a time, with small exercises you can try.

Start with your product

Your teaching prompt
Teach me PostHog AI Observability using my own product and PostHog project.


Inspect its existing PostHog instrumentation, then create a custom learning path based on its real LLM calls, traces, costs, and gaps. Teach one concept at a time and give me small, safe exercises.


Use the PostHog AI Observability docs at https://posthog.com/docs/ai-observability.md as the source of truth.


Don’t change the project unless I explicitly ask. If you can’t access the project, tell me what context you need.

Traces and generations

Your teaching prompt
Teach me PostHog AI Observability using one LLM feature in my product. Explain traces, spans, and generations, and what PostHog captures on each call. Help me find where my code calls a model, predict what a generation records, and read a real trace.


Use my code and PostHog data. Teach one concept at a time with a small exercise, and review my answer before continuing. Use https://posthog.com/docs/ai-observability.md as the source of truth.


Don’t change my project without asking. If context is missing, ask for it.

Evals

Your teaching prompt
Teach me PostHog evaluations using an LLM feature in my product. Explain why an error rate can’t tell me whether an answer is good, and compare LLM-as-a-judge and code-based evals. Help me write one eval for a real failure mode and read its pass rate.


Use my code and PostHog data. Teach one concept at a time with a small exercise, and review my answer before continuing. Use https://posthog.com/docs/ai-observability.md as the source of truth.


Don’t change my project without asking. If context is missing, ask for it.

AI cost

Your teaching prompt
Teach me how PostHog attributes AI cost using my product’s LLM calls. Explain how cost comes from tokens and the model, and which properties tie a generation to a person, a customer, and a feature. Help me find a call that’s missing those properties and break my cost down by feature.


Use my code and PostHog data. Teach one concept at a time with a small exercise, and review my answer before continuing. Use https://posthog.com/docs/ai-observability.md as the source of truth.


Don’t change my project without asking. If context is missing, ask for it.