Wayfair says 1,600 engineers now use Cognition's Devin
Release testing is one of the quieter bottlenecks in retail technology. A change to fulfilment logic cannot go live until the team is confident it will not break orders, and that confidence takes time to build. Wayfair says an AI coding agent has cut that wait sharply.
A Cognition customer case study reports that more than 1,600 Wayfair engineers use Devin and that more than 40,000 Devin-authored pull requests have been merged. It describes work on fulfilment optimisation and database migrations, including a reduction in one team's regression-testing replay time from about two weeks to about an hour.
What the case study adds
The customer case study published by Cognition gives a narrower scope for the testing improvement. It says the saving applied to the fulfilment optimisation team's order replay testing: "A replay that took about two weeks now takes about an hour". That is a big change for the team involved. It is not evidence that every regression test across Wayfair now runs in an hour.
The case study reports further gains. It says "The database migration that started in the pilot is now moving 8-12x faster." It also says the fulfilment team's "PR throughput, weighted by complexity, went up 2.78x", and that in the next planning cycle the team "took on almost 30% more work."
How far the evidence goes
These are claims from Wayfair and from the company selling the software. Neither source gives an independent measurement, and neither gives the cost of running the agent alongside the savings. Faster testing is only valuable if the quality of what ships holds up. The real test is whether release failures, customer-facing errors and incident rates stayed flat while throughput rose.
Even so, the detail is more useful than most AI announcements. Other retailers can see which jobs the agent took on: replaying orders to check fulfilment changes, and moving old databases off ageing systems. Both are slow, unglamorous jobs that sit between a commercial idea and the moment it reaches customers. A retailer weighing a similar tool could start with one bottleneck like these and measure before and after, rather than counting engineers with access.
Related coverage

Zalando adds size advice to its AI shopping assistant
Zalando said 22 million people used its Assistant this year and that new size advice lifted add-to-cart rates by 4% in testing.
Continue readingMore in AI in Retail
🔍 Explore related coverage
Five Things Friday
Five useful signals from the people shaping global retail. Every Friday.



