Loblaw Says AI Now Generates 46.9% of Its Code

Lauren Steinberg describes Loblaw’s use of OpenAI tools in blunt productivity terms: work that previously took “teams weeks and months” is now taking her “minutes and hours.” The claim is not framed as a narrow pilot. Steinberg says ChatGPT Enterprise is “in the hands of every single colleague” at Loblaw, Canada’s largest retailer, and places the deployment inside a broad digital operating environment: apparel, beauty, grocery, and pharmacy e-commerce businesses.

The customer-facing example shown on screen is a ChatGPT conversation tied to PC Express. A user asks what to cook for dinner for two adults and two picky children, requesting something simple and quick. The PC Express experience suggests cheesy chicken quesadillas, gives a 25-minute recipe, lists ingredients and steps, then asks for a postal code to find a nearby store.

After the user enters a Toronto postal code, the interface presents local store options, including Loblaws Musgrave Street, Rocco’s NOFRILLS Toronto, and Loblaws Leslie Street. Once a store is selected, the app creates a priced PC Express shopping list for the quesadilla recipe and shows a checkout path. The visible list includes BBQ chicken, tortillas, shredded cheese, salsa, and sour cream, each with a price.

That example matters because Steinberg does not describe AI only as an internal coding tool. She says Loblaw is building “next-generation consumer products” in two places: chat on Loblaw’s first-party platform and the ChatGPT PC Express app in Canada. The product ambition shown is to move from a general consumer question — “What should I cook for dinner?” — to a local-store shopping list with prices and checkout readiness.

The most concrete productivity claims concern software engineering. Lauren Steinberg says the rapid evolution of tools such as Codex has produced “incredible” developer productivity gains. The figures shown on screen are explicitly attributed to “Loblaw Internal Data,” and the article should be read on that basis: these are the company’s internal claims as displayed, not an independent audit.

The internal data card gives four specific claims: 46.9% of all code is now AI-generated; more than 70% of code is AI-generated in teams such as Digital Health; more than 30% of engineers are seeing 2x productivity, with that productivity claim described as “based on merged PRs”; and the results are presented alongside the statement “0 purpose-built tooling” and “zero coaching.”

The “based on merged PRs” qualifier is important because the displayed 2x productivity claim is tied to merged pull requests, rather than only to usage or sentiment. The “0 purpose-built tooling” and “zero coaching” line should be treated as part of the displayed internal context, not as proof of what caused the reported results.

The reaction inside the engineering organization is shown through a message from Charu Pujari: “Hey, can I show you something? This is from our best developer, who is now 10x more efficient. We are refactoring work within a single week that took us a whole year to write.” Steinberg says she sees Slack messages moving among engineering leadership with reactions such as, “these models are crazy.”

The distinction in Loblaw’s account is between broad availability of AI tools and specialized internal buildout. The displayed data presents substantial code generation and pull-request-linked productivity claims before “purpose-built tooling” or coaching are introduced as part of the story. Steinberg’s framing is that the tools themselves, including Codex, are evolving quickly enough to create visible changes in developer work.

Lauren Steinberg gives a second example outside software engineering: product imagery for e-commerce. Historically, she says, Loblaw might have had a pair of shoes photographed in only a couple of ways because that was all the company could afford “from a time or a cost perspective.”

The before-and-after visual is straightforward. The “Before ChatGPT Images” product page shows “Buckled Block Heels” priced at $39 with a single image of a brown block-heel sandal. The “After ChatGPT Images” product page keeps the same product and price, but shows four images of the shoe being worn by a model in different poses.

Steinberg says Loblaw can now show the shoes “in 10 different ways,” across different scenarios and different types of models. The claim here is not the same as the Codex claim. In engineering, the emphasis is AI-generated code, merged pull requests, and refactoring speed. In product imagery, the emphasis is expanding the range of customer-facing assets that can be produced when time and cost constraints are lower.

That changes the merchandising surface area. The old example is a limited product presentation: one or two angles of the item. The new example is a richer product page for the same item, with more contextualized images of how the shoe appears when worn. Steinberg frames this as consistent with Loblaw’s internal culture, saying the company is “nothing if not innovative” and that people are excited to have tools that let them “finally do that.”