Walmart AI agent errors have become a routine feature of store life for the retailer’s 1.6 million US workers, as the company’s push to deploy artificial intelligence across its store network runs into the practical limits of what current models can reliably do without human correction. Employees are spending portions of their shifts supplying the context, judgment and ground-truth feedback that the AI systems lack, turning frontline staff into de facto model trainers on top of their existing roles.
Sidekick and the MyWalmart app: how AI agent errors enter the daily workflow
The core tool at the centre of the rollout is an AI assistant called Sidekick, which assigns tasks and targets to employees and, according to workers who have discussed the system on Reddit, auto-assigns where people work based on their available hours. Those assignments appear in the MyWalmart app, which runs on company-provided phones and is available to download on personal devices.
Elizabeth Nigh, an asset protection manager in Wisconsin, described one genuine upside during a company-arranged store tour. ‘If this associate is really good at stocking aisle eight, and we put them in aisle eight all the time, the AI assistant learns that,’ she said. ‘Then they can just come in and look at their MyWalmart, instead of having to wait around for a plan.’ For associates who previously had to wait at the start of a shift for a manager to brief them, that represents a real efficiency gain.
The problems emerge when the AI’s model of a job diverges from how that job actually unfolds. Restocking, for example, is not simply placing new merchandise on a shelf. Expiration dates need checking, damaged goods need discarding, and surfaces need sanitising. Each of those steps takes a variable amount of time that AI-projected task durations do not consistently account for.
Safety alerts and route planning add to the burden of Walmart AI agent errors
A new AI agent that monitors store safety is also generating friction. It draws on data from security cameras and sensors throughout the store, but is still learning to interpret that data correctly. One online fulfilment worker told Business Insider she is inundated with safety alerts from other departments whenever she opens the app on her personal phone to clock in and out. Those alerts are intended to flag potential issues (spills, refrigerator temperatures drifting out of safe range) but frequently flag false positives that workers must correct or contextualise. Her response to the irrelevant ones: ‘I ignore them.’
Route planning has produced similar complaints. Drivers for the Walmart-owned Spark delivery service said a ‘smart path’ feature sometimes instructs them to collect frozen items at the start of a shopping trip rather than at the end, giving items such as TV dinners and bags of ice longer to thaw. One driver described the added steps as ‘extra time and energy I could be spending on the next order.’
An HR manager responsible for onboarding, training and scheduling put the core tension plainly, referring to the task-assigning AI’s expectations of workers: ‘It just assumes that they’re going to be perfect every single time.’ Her store has chosen not to enforce AI-generated metrics that diverge from real-world experience. ‘We’re not going to hold people to these impossible standards,’ she said.
Brooks Forrest, Walmart’s VP for associate tools, said during an interview in July that disciplinary action over AI guidance is not the company’s approach. ‘We do not punish people for not following the tech guidance,’ he said, adding that his product development process includes frequent store visits and regularly scheduled listening sessions from the company’s Bentonville headquarters. ‘The technology is there to help them, but they ultimately have the say in what they do.’
Worker survey reveals deep mistrust over AI decision-making
A survey conducted by the worker advocacy organisation Walmart-focused group United for Respect, drawing on responses from more than 250 Walmart employees this year, found widespread concern. More than 85% of respondents said they did not trust Walmart to prioritise their needs in developing AI, though roughly 40% acknowledged that AI could make their job easier and a similar share saw potential for it to eliminate repetitive tasks.
Nearly three-quarters of respondents identified AI involvement in HR decisions (wage reviews, time-off requests) as their top concern. Two-thirds worried about reduced contact with managers and colleagues. Half feared being penalised for mistakes flagged by AI.
Ava Williams, an overnight stocker in Spokane, Washington, presented a United for Respect shareholder proposal at Walmart’s shareholder meeting last month asking the company to produce a report on AI’s impact on its workforce. She said AI-directed workflows push her team to cut corners on sanitising shelves or checking for expired products just to keep pace. ‘We’re expected to meet impossible timelines,’ she said. Walmart’s board opposed the proposal, arguing that technology should serve workers by ‘elevating human capability, improving the customer experience, and making work more meaningful.’ The measure was not approved, but United for Respect’s filing means the question of how Walmart accounts for AI’s workforce impact is now formally on the record with shareholders.