:quality(75))
Adobe announced Agent Orchestrator on March 21, 2025, as part of Adobe Experience Platform, along with 10 specialised AI agents. The idea is simple enough: instead of one agent fielding every request, work gets handed to agents that each do one thing.
How it works
Agent Orchestrator is a framework for getting multiple AI agents to work together on the same job. One agent handles inventory queries, another manages returns, another optimises pricing. They communicate, share context, and execute decisions with minimal human intervention.
The 10 agents
The initial 10 cover:
customer service — answering questions and escalating issues
inventory management — stock allocation and replenishment
order processing — fulfilment and tracking
returns management — processing refunds and exchanges
pricing optimisation — dynamic pricing based on demand
personalisation — recommending products based on customer history
fraud detection — identifying suspicious transactions
customer loyalty — managing rewards and promotions
content management — updating product information
analytics — generating business insights from sales data
Orchestration in practice
Take a customer ordering something that has just gone out of stock. The orchestrator picks up the inventory problem, asks the inventory agent to check availability at other facilities, coordinates with the logistics agent on alternative shipping, tells the customer service agent what has happened, and updates the customer. No one touches it.
The agents also learn from outcomes — which recommendations convert, which shipping options reduce returns, which pricing strategies hold margin — so the decisions shift over time.
The bit that needs a decision
Agent Orchestrator needs clear rules on agent authority: where an agent can act on its own, and where it waits for a person. Most merchants land on a hybrid — agents handle the routine and escalate the exceptions. That line is yours to draw, and it is worth drawing before you switch anything on.
What this means for you
If returns processing, order chasing and customer research currently eat your team's hours, that is the work this is aimed at. The question isn't really whether the agents can do it. It's which decisions you are willing to let them make without asking first.