About 12,000 Shoprite Group employees are using generative AI tools, and group chief technology officer Chris Shortt says the retailer may have to plan for a future in which new recruits turn up with their own AI agents. He has raised it with the group's chief people officer, including whether a token allowance might form part of a remuneration package.

There is no policy behind it, and Shortt framed it as internal thinking rather than anything decided.

The detail underneath is more useful than the headline idea. Shoprite built its own platform, Shoprite AI, which presents staff with a ChatGPT-style interface while gateways behind it route each request to whichever model can handle it most cheaply and effectively. Shortt is blunt about why: if you do not watch the economics of these things, they can get out of hand very fast. That is least-cost routing applied to model selection.

On the engineering side, he has placed no constraints on how his teams use AI and is actively encouraging it, and the whole software development lifecycle has compressed, starting with how requirements and business analysis get defined before anything reaches an engineer. His teams deliberately play models against each other, working from a specification with Claude Code, then cross-checking against Gemini, on the basis that each finds something the other missed.

A delivery team now typically looks like one or two humans plus several agents building alongside them. The 12,000 figure sits against a workforce of more than 174,400, most of it in stores.

TechCentral reported Shortt's comments from its Meet the CIO podcast, and notes how unusual this level of operational detail is from a listed company executive. The full interview covers what Shortt thinks a supermarket becomes. Shortt joined Shoprite in January 2024 after more than 12 years at UCS Solutions and a spell as group chief information and technology officer at Pick n Pay.

Two things here are worth acting on rather than admiring. The first is the cost gateway. Any business running AI at scale eventually discovers that routing every query to the most capable model is a choice it cannot afford, and Shoprite built the plumbing to avoid that before the bill arrived.

The second is Shortt's advice to anyone starting, which is to build a skills base that puts you in a position to tell the machines what you need and then validate what they produce. He calls these probabilistic models rather than deterministic finalities, and says validation stays with the person using them.

That is the answer to the question hanging over developer career frameworks and over the junior roles the Digital Skills Impact Fund is co-funding. The entry-level task is disappearing. The judgement to check the output is not, and Africa's largest retailer is now hiring for it explicitly.

You might also like our piece on how Pixie Smart Chat handles measurement inside Sixty60, what the Cognition valuation says about agents doing the grunt work, and why SA enterprise AI adoption keeps outrunning its own strategy.

Get more SA tech and business news and subscribe to The Open Letter.