Industries / Retail, hospitality and personal services / Retail and e-commerce
Online shops and search
Online selling is selling at a distance, from printed catalogs to websites and apps, with search and recommendations to help shoppers find things, and now AI agents that shop for them.
6
Milestones achieved, from 1872 to 2026.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1872
Mail-order catalog opens
Montgomery Ward, in Chicago, began selling from a catalog of more than 100 products, so families far from a store could choose goods and order them by mail. Sears, Roebuck and Company followed in 1893. The US Postal Service calls Montgomery Ward "the first mail-order company."
Credit
Aaron Montgomery Ward; Sears, Roebuck and Company
1984
Groceries ordered by TV
In May 1984 Jane Snowball, aged 72, used a modified television and a remote control in her home in Gateshead, England, to order groceries from her local Tesco. The order went down her phone line, and the shop delivered it. The scheme was run by Gateshead Council for older people who had trouble getting to shops, using a system built by Michael Aldrich. The BBC calls it "arguably" the world's first home online shopping.
LessonIt worked but never spread. The BBC quotes the view that with only a few customers it is "extra labour for not much extra profit." Home shopping online waited for home computers and the internet.
Credit
Gateshead Council; Michael Aldrich; Tesco
1995
A bookstore on the web
Amazon.com opened in July 1995 as "Earth's Biggest Bookstore." Shoppers searched a catalog of more than 2.5 million titles instead of walking shelves. Its 1997 stock-offering prospectus says it had sales to about 340,000 customer accounts in more than 100 countries by the end of March 1997.
Credit
Amazon.com (Jeffrey P. Bezos)
2003
Recommendations that scale
Amazon engineers published how its "customers who bought this item also bought" feature works. The method, item-to-item collaborative filtering, links each item a shopper bought or rated to similar items instead of searching for similar customers, so the work done while a shopper waits does not grow with the number of customers or products. In 2017 the journal IEEE Internet Computing chose it as the paper from its first 20 years that had best "withstood the test of time."
Credit
Greg Linden, Brent Smith and Jeremy York (Amazon.com)
2020
Lockdowns push shopping online
When stores closed during COVID-19, the US Census Bureau measured online sales rising from 11.8% of US retail sales in the first quarter of 2020 to 16.1% in the second. In Great Britain the Office for National Statistics measured a rise from 19.1% of retail sales in February 2020 to 37.8% in January 2021.
LessonMuch of the jump did not last. The British share was 28.3% in September 2026 and the US share 17.1% in the second quarter of 2026, both still above where they started.
Credit
the US Census Bureau and the UK Office for National Statistics (measurement)
2026
Chatbot checkout pulled back
In September 2025 OpenAI and Stripe launched Instant Checkout, which let US ChatGPT users buy from Etsy sellers without leaving the chat. It ran on the Agentic Commerce Protocol, an open standard for letting AI agents place orders. In March 2026 OpenAI let merchants send buyers to their own checkouts instead and turned ChatGPT shopping toward helping people find products.
LessonRetail Gazette reported that the problems were ordinary retail work: "merchant onboarding, maintaining accurate product data, enabling multi-item baskets and integrating loyalty schemes.".
Credit
OpenAI and Stripe
Watch for
Checkout inside Google's AI search
Google announced the Universal Commerce Protocol, an open standard for AI agents to handle a purchase from search to checkout, co-developed with Shopify, Etsy, Wayfair, Target and Walmart. Google says shoppers can check out from eligible US retailers inside AI Mode in Search and the Gemini app.
Achieved whenretailers or an independent measurer report how many sales actually go through agent checkout, and the service runs outside the US.
Claimed by
Google, January 2026
Amazon's shopping assistant at scale
Amazon says its "agentic shopping assistant", Alexa for Shopping, "was used by 400 million+ customers in the last 12 months, with customer interactions up 570% year-over-year."
Achieved whenan independent party measures the use and how much it changes what people buy.
Claimed by
Amazon, third-quarter results, October 2026
Outside agents shopping store sites
Perplexity says its Comet browser agent may shop on Amazon when a user tells it to. Amazon sued in November 2025, and on 9 March 2026 a federal judge in San Francisco blocked Comet from Amazon. In August 2026 a federal appeals court overturned that order, and the case continues.
Achieved whena final judgment, settlement or law settles whether retailers may block agents that act for a user.
Claimed by
Perplexity, November 2025 onward
Open
01
Pack facts on grocery websites
A checker that compares an online grocer's product page with the label on the actual pack (ingredients, allergens and nutrition facts) and reports which facts are missing or different, store by store.
The briefCredit · unclaimed02
Blind shoppers test online checkouts
A published, repeatable test in which blind screen-reader users try to buy from large online shops, up to the payment page, and an AI agent that drives a screen reader and predicts where they will get stuck.
The briefCredit · unclaimed
Open 01 · the brief
Pack facts on grocery websites
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A checker that compares an online grocer's product page with the label on the actual pack (ingredients, allergens and nutrition facts) and reports which facts are missing or different, store by store.
Why it is still open
A study by Tufts University and New York University, published in Public Health Nutrition in October 2024, looked at 60 food products sold by 10 online grocery retailers. Each label that the US Food and Drug Administration (FDA) requires was "present, accessible, and legible for just 35.1% of products." US law puts these facts on the package, and online retailers are not required to repeat them.
Built with what exists
pack photos from Open Food Facts; Playwright or Browser Use (see the Note "Deleting 80% to start over") to save product pages, at a slow rate and within each site's rules for automated visitors; vision-language models to read labels; Python.
First step
Pick 100 products with clear label photos in Open Food Facts, and record by hand what three online grocers show for each.
Done looks like
2,000 products across five US online grocers, each checked by the tool and, for a random 200, by two people. The tool agrees with the people on at least 95% of missing or mismatched facts, and a table by retailer is published.
Open 02 · the brief
Blind shoppers test online checkouts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A published, repeatable test in which blind screen-reader users try to buy from large online shops, up to the payment page, and an AI agent that drives a screen reader and predicts where they will get stuck.
Why it is still open
WebAIM's 2026 scan of the top 1,000,000 home pages found detectable failures of the WCAG 2 accessibility guidelines on 95.9% of them, and shopping home pages averaged 71.0 errors each. WebAIM notes that "not all conformance failures can be automatically detected." The European Accessibility Act covers e-commerce services provided to consumers after 28 June 2025. A search found commercial tools that simulate screen readers but no public tally of blind shoppers completing checkouts across major shops.
Built with what exists
the free NVDA screen reader on Windows and VoiceOver on Apple devices; Playwright to record each step; an AI agent model that reads the screen reader's output instead of the screen.
First step
Recruit three blind testers, run the test on five shops, and publish the method.
Done looks like
50 large EU online shops, three blind testers each, adding two items to the basket and reaching the payment page. The published tally shows completions and each blocking step, and the agent's predicted blockers match the testers' on at least 80% of shops.