Most analytics tools tell you what happened. James Evans argues that in the age of AI, tools need to tell you what to do next.
In this episode, James shares how Amplitude is rethinking analytics with AI -- from anomaly detection that surfaces fixes, to session replay that could replace brittle event taxonomies. We dig into the cultural and technical shifts teams need to make to become AI-ready: embedding copilots into real workflows, running hack sprints instead of hackathons, and proving adoption through prototypes, not PRDs.
—
Brought to you by Autoskills (https://autoskills.ai) – Helping teams go from AI-curious to AI-ready with tailored workshops and hands-on adoption.
—
What You’ll Learn
—
What’s Covered
(00:00) James' introduction and the Command Bar acquisition
(09:27) How to build background agents that find signals and trigger fixes
(10:21) Prototype a single onboarding agent to spot drops and prove impact
(13:49) Why you need to apply LLM filters to cut alert noises
(15:06) How to treat session replay as your causal X-ray for true issues
(20:59) Lock down eval metrics before you scale generated UIs
(28:37) Pull instrumentation bugs from sessions and push actionable fixes
(29:43) Pilot session-first pipelines to avoid brittle taxonomies
(31:41) Pair conversational queries for discovery with a GUI for deep analysis
(36:31) Seed adoption by solving internal time-suck problems with copilots
(54:10) Quick tips on how to win at your AI project
—
Where to Find James Evans:
Where to Find Haroon Choudery: