The Skift Takeaway
Skift Data + AI Summit 2026
Should You Be Deploying Agents Now or Are You Already Late?
Pol Peiffer, Head of Product and Agent Development, Sierra · Gaëlle Bristiel, Senior Vice President of Engineering, Amadeus
Moderated by Vivek Bhogaraju, Executive in Residence, Private Equity
THE ARGUMENT
Peiffer said agents are production-ready now and that the industry’s demo-land problem is a choice, not a constraint. Bristiel argued the harder problem is ecosystem fragility: without sufficient traveler context shared across systems, even strong agents can return irrelevant results. The session’s real tension wasn’t whether to deploy but what preconditions make deployment succeed — and whether companies that wait for those conditions to be perfect will simply run out of time.
THE EVIDENCE
- Peiffer cited an MIT study that found roughly 90% of AI deployments were getting stuck in demo mode; Sierra sees the opposite, with over 95% of pilots reaching production.
- Sierra worked with an APAC airline to build an agent, which launched in April; it supports English, Tagalog, and “Taglish.” Sierra aims to grow containment — contacts handled without human intervention — from 30% to 50–80% over time.
- “The important part is the pace of iteration and improvement, not necessarily your starting point.” — Peiffer
- A Bristiel-led project failed when an AI trip-recommendation system returned results rejected as irrelevant because the customer shared too little context about the traveler’s situation and history.
- “You don’t need to wait another year or two, you’ll probably be behind by then.” — Peiffer
THE SO WHAT
If your organization is still waiting for agents to mature before production deployment, this session argued the waiting period is already costing competitive ground. The practical precondition isn’t technical readiness — it’s defining the end-to-end outcome you need and ensuring the data required is actually shared across your systems.
