Thank you to everyone who attended yesterday’s Watson Live! — Agentic Debate Series Lunch @ Shoptalk Las Vegas 2026, presented by Logicbroker event. The event video will be on the Watson Weekly YouTube channel in a few weeks’ time.

The conversation I didn't expect to have this week was about customer support. It turned out to be one of the sharpest 25 minutes I've spent thinking about where e-commerce operations are actually heading — not where the press releases say they're heading.

Darren Heaphy runs product at eDesk. He has worked in logistics and e-commerce long enough to have watched multiple technology shifts arrive, get overhyped, and eventually settle into something useful. Here is what stuck.

A seasonal hire makes mistakes in week one. Brands accept it, train through it, and move on. The same brand deploys an AI agent, it mishandles 3 tickets, and the whole program gets pulled.

Heaphy calls this the AI double standard — and the diagnosis is precise. The fear is not really about error rates. It is about visibility and scale. A human mistake feels contained. An AI mistake feels like it is happening everywhere at once, invisibly, and that perception is driving decisions that the data does not support.

The correction is not blind trust. It is the same graduated trust you would extend to any new team member: start with low-risk work, build an audit trail, expand scope as it earns it.

This number stopped me. Between a 30% and 40% of inbound support volume is pre-purchase inquiry — sizing questions, return policy questions, availability questions. Not complaints. Buying signals.

Most brands route those tickets into the same queue as post-purchase problems and handle them with the same urgency, which is to say, not much. An AI agent working that queue around the clock, trained to recognize purchase intent, converts it. The support desk stops being a cost center and starts producing revenue.

The infrastructure for this already exists. The reframing is the hard part.

The future Heaphy describes is not a support team of zero. It is a support team that stops typing and starts deciding. Humans become the conductors — training models, managing exceptions, and reviewing the cases AI flags as outside its confidence threshold.

The physical act of composing a response to a routine inquiry is going away. The judgment required to handle escalations, edge cases, and anything that requires context a model cannot hold — that is not going anywhere.

The brands that understand this distinction will staff and train accordingly. The ones that do not will either over-invest in automation or under-invest in the human layer that makes automation trustworthy.

Heaphy closed with a line worth keeping: people feeling overwhelmed by AI right now are not behind. They are paying attention.

That is true.

Paying attention is not a strategy. The brands that will come out of this transition in better shape are the ones treating AI adoption the way they would treat any new hire — with structure, with accountability, and with a clear path from low-risk tasks to high-trust decisions.

The technology is not the hard part. The management is.

Chewy (After market close, 23 March 2026): Q4 will show whether the Q3 momentum held through the holiday period. The Autoship attach rate and active customer count are the two numbers worth watching