- AGS Events
- Meetups
- News
- Magazine
- Media Partners
- About
Dayu Yang used Codi, his storytelling toy, to explain AI systems at AGS Rome 2026 on 05 November. The EcommOps CEO launched the voice-enabled toy in 2018, adapting its speech recognition for young children. His Day 3 keynote was Beyond Vibes: Building AI Systems That Give Your Ecommerce Business a Lasting Edge. He organised the talk around four layers: use cases, data, process and people.
Yang warned that an assistant could misread inventory if its information did not reflect physical stock. “AI vibes do not make a defensible business, and most likely, without building in the right way, you’re creating slop.” His example involved an assistant predicting a three-week supply of stock before the business sold out two days later. The problem was its reliance on order data without knowing whether those orders matched real inventory.
Codi needed to work for two different audiences. Children aged two to five needed a soft, engaging toy with speech recognition adapted to how they talk. Parents wanted an app to control the device and see how their children used it. The same product therefore required two different experiences.
That distinction also applies to inventory and order tracking. An inventory dashboard needs to account for lead times, sales projections and when new stock must arrive. A customer service team needs detailed shipment information, particularly around errors and delays. Customers need a simpler tracking page that makes their order’s progress easy to understand.
Useful AI systems depend on reliable and connected data. Yang described a client recreating a tracking page with AI to avoid paying for an existing service. The page looked right, but nothing appeared because it had no connection to tracking data. Building the page had not replaced the information feeding it.
At EcommOps, warehouse records map physical locations down to rows, shelves and trays. Every inventory movement links to a physical scan. Brand databases consolidate order, inventory, tracking, and other operational information before dashboards or AI tools use it. Yang stressed managing that data separately from AI outputs, so errors do not become buried in the underlying records.
Reliable information also needs a reliable sequence of actions. EcommOps uses at least 10 steps per order, tied to physical scans. Staff cannot move to the next stage until they complete the previous one. Tracking numbers go out at the final stage, after fulfilment, rather than before stock is ready to ship.
Courier updates create another processing problem, with almost 600 tracking event types arriving from different providers. Code reduces those events to a smaller and more understandable set for staff and customers. Yang described using AI to build that code, then running it without AI. The distinction is important: AI can build a tool without powering every action it performs.
People sit at the foundation of Yang’s AI systems. Codi brought together a technologist, an educator and a toy maker. At EcommOps, staff move inventory, pack orders and check the work at each stage. Yang also described human strategy and review within a 16-step process for creating static adverts.
Those workflows need to fit the business and the task at hand. The process for static adverts differs from the process for video, even within the same brand. Yang argued that the competitive advantage comes from customising all four layers together. No single layer provides that protection on its own.
Yang ended with a warning about who should remain in control. “I don’t want human beings looking outside in at AI and robot monsters taking over the world. So think about the system in this way. Build the system, manage the system, but do not let the system replace you.”
TrafficStars supports AGS Rome 2026 as main sponsor. It connects advertisers with large-scale premium inventory across global markets, offering both self-serve and fully managed campaign options.
Join us for the final day of AGS Rome 2026 on 05 November. Visit Affiliate Grand Slam for today’s programme and event details.