Mobile Dev Memo Podcast Season 7, Episode 9: RecSys and internet commerce (with Michael Komasinski)
9 snips
Mar 11, 2026 Michael Komasinski, CEO of Criteo, leads the company through commerce and adtech innovations. He explores agentic commerce and Criteo’s role as a commerce intelligence layer for AI assistants. They compare recommendation systems to large language models, discuss the OpenAI advertising partnership, Criteo GO self-serve ads, and how recommendation tech reshapes product discovery and retail integrations.
AI Snips
Chapters
Transcript
Episode notes
Commerce Intelligence Layer Beats LLMs For Recommendations
- Criteo built an agentic commerce recommendations service as a commerce intelligence layer for AI assistants.
- Offline A/B tests show ~60% uplift in product relevance by using real-time purchase signals and purpose-built recommendation models.
LLMs And Rexus Serve Different Roles
- Large language models are semantic and excel at language, while recommendation systems (Rexis) need reward loops and high-volume outcome signals.
- Criteo positions its backbone as a Rexus platform and expects future systems to be hybrids of semantic and Rexus models.
Design For Assisted Agents Not Fully Autonomous Buying
- Avoid assuming full autonomy for commerce agents; design for augmented or assisted experiences where users retain final approval.
- Criteo focuses product roadmaps on surfacing recommendations while preserving user agency and multiple touchpoints.