Agentic commerce and retail, three notes on who decides the purchase
Agentic commerce is no longer a hypothesis. In the notes below, the author walks through an autonomy curve for purchasing, the growth of agent-originated searches at a British retailer and what a beauty case shows about operating AI at scale. Each note carries its original date and a link to the post. The data is valid as of the date of each note.
How much of the purchase decision we will delegate
February 13, 2026
Agentic commerce is not a trend. It is the next layer of automated decision-making, and it is already operating. McKinsey published a curve with six levels of autonomy, from programmed convenience, such as automatic subscriptions, to networked autonomy, with agents negotiating with other agents.
The most important point is that not everything will be level 5. Recurring purchases evolve fast. High-value or emotionally charged decisions will keep meaningful human supervision. And autonomy without governance becomes systemic risk.
The technology already allows delegation. The strategic question is whether we are ready to structure governance before automating.
Source, The automation curve in agentic commerce, McKinsey, January 28, 2026 · Original post
The next shelf will be read by an agent
September 4, 2026
Retail's next shelf will not be seen by the consumer. It will be interpreted by an agent. At John Lewis, product searches originating from AI agents jumped from 0.3% to 2.5% in a year, according to Reuters.
When agents research and recommend, a new demand channel emerges with its own funnel, from presence in answers to incremental revenue. My experience taught me a clear lesson. Visibility without incrementality can become just a more sophisticated vanity metric.
We may move from an obsession with clicks to a new obsession with mentions in AI answers, without knowing whether any of them produced a sale that would not have happened otherwise.
Source, Reuters, September 3, 2026 · Original post
Operating AI at scale, the L'Oréal case
September 24, 2026
L'Oréal did not put generative AI in a chatbot. Beauty Genius combines generative models, RAG on proprietary knowledge and dedicated infrastructure, and the sophistication lies behind the scenes, in an architecture built for demand peaks.
There is an important caveat. Public data proves technical performance and the ability to scale, but does not yet demonstrate conversion, revenue or ROI.
The portal published a story on this case and on Crocs, covering what each company disclosed about scope and reuse.
Read Future in Use's story on L'Oréal and Crocs · Original post
Further reading
- AI agents on WhatsApp, what the published Agibank and Riachuelo cases describe
- The enterprise knowledge layer, the context AI agents need to operate
- Gustavo Gadotti maps in five blocks what AI agents change in retail, from the cart to the sales team
Notes originally published on LinkedIn between February 13, 2026 and September 24, 2026, with each post's link next to its note. This collection has been edited and organized with context and sources. The views are those of the author, a co-founder of OperOS AI.