On September 17, 2026, Agnic.AI joined an ALL IN plenary on the Forum Stage in Montréal to examine how agentic commerce could reshape purchasing, product discovery, and customer relationships. April Fong, AI and Tech Editor at The Globe and Mail, moderated the discussion with Alix Moghadam, Founder and CEO of Agnic.AI; Tom Pawelkiewicz, VP, Digital Payments at Mastercard; Lauren Steinberg, EVP and Chief Digital Officer at Loblaw; and Louis Têtu, Executive Chairman of Coveo. Trust emerged as a central condition for adoption, with the panel exploring the controls, expertise, and accountability that autonomous transactions require.
Early transactions, substantial expectations
The panel drew a clear distinction between industry activity and actual adoption. Steinberg reported that Loblaw’s online grocery site had recorded only three agent-made transactions in 30 days, despite a few hundred agent visits. Pawelkiewicz described the market as not yet commercially meaningful and identified 2028 as a potential breakout year. These were observations and expectations shared on stage, rather than evidence that broad adoption has already arrived.
For Agnic.AI, that gap puts trust at the centre of the adoption question. Moghadam argued that customers should not simply hand payment credentials to an AI today. Greater autonomy depends on infrastructure that establishes identity, proves authority, and keeps people in control.
The hard part is the control layer
Têtu described an agent as a model plus a “harness”: the permissions, intent, and business rules that govern its behaviour. A retailer may need an agent to prioritize a seasonal promotion, protect margins, or clear inventory. As models become more interchangeable, the controls around them remain specific to each business.
Pawelkiewicz emphasized verifiable intent: establishing that a user was authenticated and that the purchase matches what the user authorized. He also distinguished an immediate instruction to buy from a future mandate, such as purchasing an item only when its price falls below a specified threshold. The latter requires authority to persist over time while remaining bounded.
Moghadam’s contribution connected these controls to agent identity and delegation. A mandate can begin in plain English, but its meaning must be understandable across the agent, merchant, bank, and payment processor. Agnic.AI’s view is that businesses may adopt delegated agent action before retail consumers because business systems already define roles, permissions, and approval boundaries.
Retailers need to remain experts at the moment of intent
Steinberg discussed Loblaw’s PC Express experience in ChatGPT, including the ability to turn a recipe into a grocery basket. The example exposed a division of expertise: an AI platform can help interpret a request, while the retailer holds the local knowledge of live inventory, pricing, and fulfillment needed to deliver it.
The panel also challenged the idea that personalization is already solved. Steinberg argued that current agents often make generic selections and miss individual taste. The discussion distinguished basic recommendations from understanding a product in the customer’s specific context. Delegating a routine replenishment purchase is a different proposition from choosing travel or a deeply personal item.
Steinberg’s proposed direction was an “agent for agents”: a retailer-owned expert that can interact with outside assistants. Têtu similarly argued that brands should remain present at the moment of intent. Together, these perspectives suggest that agentic commerce will require retailer expertise and transaction capabilities to be accessible through new interfaces.
Discovery moves from keywords to meaning and reputation
All four panelists indicated that they expect consumers to visit retailer websites less often over the next decade. Their recommendations nevertheless began with work merchants can do today: improve product information, make capabilities discoverable, and provide a reliable experience when an agent arrives.
Têtu highlighted thin catalogs and weak product semantics and taxonomy. Pawelkiewicz recommended describing products in the language of a customer’s situation and managing reputation beyond a merchant’s own channels. Steinberg cautioned against cluttering human-facing pages with text written solely for agents, favouring dedicated retailer expertise that other agents can access.
Moghadam framed agents as a new customer layer on top of existing e-commerce. Discovery increasingly depends on meaning and reputation, extending beyond the keywords and backlinks associated with traditional search. He also raised the need for decentralized reputation infrastructure to which agents can contribute. For small businesses and independent builders, being discoverable was Agnic.AI’s most urgent recommendation.
Accountability requires evidence and user control
Liability produced a substantive difference of emphasis. Têtu placed responsibility with the brand and business. Pawelkiewicz stressed tracking intent throughout the journey so a failure can be traced to its source, whether that is the user, agent, or fulfillment process. Moghadam argued that responsibility cannot sit entirely with brands: users also need tools to delegate authority and remain in charge of their agents.
This makes an auditable record of authorization and execution more than an operational convenience. It helps distinguish what the user requested, what the agent was permitted to do, what was purchased, and where an error occurred. Clear limits and revocation controls support that accountability before a dispute arises.
Practical investments before full autonomy
The closing round focused on discoverability, dependable checkout, and useful workflows inside existing channels. Steinberg described a flu-shot booking flow that had moved from an 11-page form to a short prompt-based interaction taking under a minute. Her recommendation was to build agentic workflows within a company’s own channels now, making them available to outside agents later.
Loyalty was another opportunity: agents could help customers understand points, balances, and offers, making benefits easier to use. The panel also discussed Canadian sovereignty. Têtu emphasized governance and resilience, while Steinberg described designing systems so underlying models could be replaced, including with Canadian alternatives. These themes reinforced the value of retaining control over the business experience as the technology evolves.
Implications for agentic commerce
The central implication for Agnic.AI is that capability and permission must develop together. An agent’s ability to complete a task does not establish its authority to do so. Deployments need a clear way to verify whose authority an agent uses, how far that authority extends, and how it can be withdrawn.
Agnic.AI’s work on identity, delegation, and payments addresses this foundation for agentic commerce. The ALL IN discussion connected that infrastructure focus with the practical concerns of retailers, payment networks, and AI providers: earning customer trust, retaining business expertise, and making agent actions accountable.
Earlier in the week, Agnic.AI participated in Forum Fintech Canada. Read the Forum Fintech Canada report.