An AI Product Manager (AIPM) is a product manager who builds, prototypes, and ships software directly with AI agents like Claude Code, using semantic search, repos, and evals as their daily toolkit. The role has shifted from writing specs and managing handoffs to chatting with codebases, vibe-coding prototypes, and designing evaluation loops for stochastic outputs. This page explains the AIPM workflow, the tools that matter, and how PMs at companies like Lemlist restructured their job in under three months.
Audience: Product managers, heads of product, and founders re-tooling their PM function around AI-native workflows.
Chat with codebase is a PM workflow where product managers run Claude Code against their company's entire repository and ask questions to understand how years of software has compo…
A vibe-coded internal tool starts as a one-to-two-day prototype built in Lovable or Replit to test functionality with the user. Once it shows fit, an MVP is developed and handed to…
A harness is the system around a large language model—planning, context, evaluations, and tooling—that turns raw intelligence into useful work. Practitioners report that models hav…
Semantic search, repos, and evals are the three primitives of modern AI product work. Semantic search retrieves by similarity instead of exact IDs, repos store reusable context lik…
Value stream mapping for AI means sitting with everyone in a process — from sales through invoicing, implementation, and customer support — to map what they do and which systems th…
Vibe coding is a workflow where product managers prototype features by chatting with AI agents like Claude Code instead of writing specs and waiting on engineering. At Lemlist, PMs…
Elements Agents ships AI workflows in production for operators in this space. Fixed-price Diagnostic, four-week Sprint, monthly Accelerator.