# How Wayfair's 1,600+ engineers went all in on cloud agents

1,600+

Engineers using Devin

40,000+

Devin-authored PRs merged

2 weeks to 1 hour

For regression testing

8-12x

Faster database migrations

2.78x

PR throughput on the fulfillment optimization software team

## About the company

[Wayfair](https://www.wayfair.com/) sells everything for the home to about 21 million active customers, with more than $12 billion in annual revenue, a catalog of 40 million items, and a supply chain built to move large, bulky products quickly.

Industry: eCommerce**[Visit site](https://www.wayfair.com/)**

## Why home is hard to sell

Fiona Tan, Wayfair's CTO, thinks home may be the hardest category to sell online. Wayfair is determined to make it easier, from helping shoppers discover and find products, to checkout, to delivery.

One part is matching. People care a lot about how their home looks, and they often can't describe what they want until they see it. Wayfair carries 40 million items so every shopper can furnish their home the way they want, across every style and price point. It has to help each shopper find the pieces that are right for them.

> "How do we make it easy and delightful for customers to browse, get inspired and find something they really love? And how do we help them feel confident that it’s going to work in their home."
>
> — Fiona Tan, CTO, Wayfair

Another is fulfillment. Many of those items are big and bulky, and they have to reach a customer's home quickly and without incident. When an order comes in, Wayfair's systems pick where the item ships from, whether directly from a supplier or from inventory housed in one of Wayfair's warehouses, whether it is delivered by Wayfair or a delivery partner, and how long delivery should take, based on large amounts of order and shipping data.

> "There's actually a lot of complexity under the hood from checkout to delivering the perfect pink polka dotted bar stool into your home as quickly as possible."
>
> — Fiona Tan

## Ambitions backed by software

Every step of that experience runs on software, and Fiona wants to shorten the time between identifying an improvement and delivering it to customers.

Amanda Holland's fulfillment optimization software team is one example. It builds the systems that route every order, so the team holds a high bar: before shipping a change, it replays past orders through the new logic and checks that the result is the same or better. Gathering the data and running a replay took about two weeks per release.

Database work is another. Wayfair wanted to move its core databases off on-premises SQL Server, in instances of up to 30 TB, and the project had been stalled for four years.

## Victor's weekend with Devin

Victor Andreoni Paseka, Director of Software Engineering at Wayfair, led the company's move to Devin and now leads its Devin Center of Excellence. Victor had been trying cloud agents since late 2025, but hadn't been satisfied: he found they required too much back and forth, and he didn't trust their outputs.

In March, someone told him to give Devin a second shot because things had changed. He signed up on his personal account on a Friday. The next time he looked up, it was Sunday, and he had spent the whole weekend on it.

His next question was whether Devin would work for a team of 1,500 engineers.

> "I could go as fast as I could think of things that I wanted to do... I realized, I need to bring this to Wayfair."
>
> — Victor Andreoni Paseka

## Going all-in on cloud agents

Wayfair's engineers already had AI coding tools, but Victor kept seeing them hit the same two walls. They couldn't easily connect to the internal systems Wayfair's code depends on, and because they ran on each developer's local machine, each one could only work on 1-2 tasks at a time.

> "We still felt that we could go faster, so we decided to pursue remote agentic as a paradigm shift for us."
>
> — Victor Andreoni Paseka

Second, they could connect to cloud-hosted SaaS, but couldn’t connect to private and self-hosted systems in their internal network.

Victor and Fiona began to see cloud agents as the foundation for how Wayfair would build from then on.

> "We felt cloud agents were going to be a big part of how our engineers work. We wanted to give our teams the best tools we could, and the opportunity to see what they could do with them."
>
> — Fiona Tan

Victor ran a bake-off of several platforms by opening them up to engineers and collecting their feedback. Devin was the easiest to connect to Wayfair's systems, and it could reach anything a developer could. With other cloud agents, each engineer spent a week or two setting up their own workspace, and those setups didn't always stay stable.

> "Each engineer has to set up their own VM or be able to create their own workspace. Whereas with the Devin cloud agents, we really felt like we were just up and running from the beginning."
>
> — Amanda Holland

## From a 20-person pilot to 1,600 engineers

Wayfair planned a pilot of 20 engineers. Cognition asked Wayfair to bring its hardest problems, so the pilot included the stalled database migration.

As engineers saw what their peers were doing, they asked to join, and the pilot grew to 100 people across web, apps, backend, and frontend.

Wayfair decided to open Devin to every engineer. Cognition's team spent a day in Boston with about 20 Wayfair platform and infrastructure engineers to work through every remaining integration and connect Wayfair's Okta groups to Devin. Wayfair mapped its org into Devin, down to more than 200 small teams, so each team's sessions start with that team's context, and Wayfair's coding rules load into every session. Before Devin, there was no standard setup, and engineers traded AI tool configurations in Slack. Now a new engineer can log in and ship a first PR within an hour or two.

Cognition's forward deployed engineer, Sam Fertig, worked with Wayfair's teams from the start of the pilot. Engineers took questions straight to Sam and the Cognition team in Slack. When Wayfair needed something Devin didn't have yet, like Android and iOS simulation, Cognition helped teams find workarounds so they could keep moving, then built it.

> "When someone has an issue with the platform, we don't have to put it in a ticket and wait. Instead, we can just Slack the Cognition team and they get back to us very quickly. They have become part of the team in that sense."
>
> — Victor Andreoni Paseka

Wayfair planned for 500 engineers. More than 1,600 now use Devin, and they've merged more than 40,000 PRs with it.

> "We were surprised at how much pent up demand that was for it, and the word of mouth on how good it was to use."
>
> — Fiona Tan

## How engineers work now

With Devin, Wayfair's engineers changed how they work. They now spend more time on design and architecture up front, then hand Devin well-scoped tasks and run many sessions in parallel. What comes back is a PR that has already been reviewed and tested.

> "Having a wider loop has been a game changer for us... You give it a plan, you get back a PR that has already been reviewed, that has already been tested."
>
> — Victor Andreoni Paseka

## Shipping faster without adding regressions

Wayfair wanted to ship faster and keep the same operational rigor. Each Devin session runs on its own machine with Wayfair's dependencies installed, so Devin builds a change, runs it, and tests it before an engineer looks at it. For the apps, Devin runs Android and iOS simulators and clicks through the app to check that nothing broke, and teams can go from a PR to a live demo of the change.

## Work that starts on its own

Wayfair also set Devin up to start work on its own. Automations kick off sessions from a Slack message, a webhook, or a schedule, and teams use them for incident response, end-to-end and integration testing, ticket triage, and backlog grooming. About 20% of Wayfair's Devin use now runs through automations, and automated testing is up 9.5x for teams leveraging this capability.

## Cutting regression testing from two weeks to one hour

The fulfillment optimization software team used Devin to evolve the testing environment that pulls historical order data and replays those orders through the new fulfillment logic on its own machine. A replay that took about two weeks now takes about an hour, and the team knows before a release whether a change will have the effect it should. Amanda doesn't think they would have found time to build it without Devin.

## Taking on 30% more work in one planning cycle

Amanda's team plans its work in six-month cycles, and before Devin, new requests from the business could push a planned project out. In the cycle after they adopted Devin, they finished everything they had committed to and took on almost 30% more work. Their PR throughput, weighted by complexity, went up 2.78x, and one engineer can now own a project from triage through deployment.

> "This is the first time in my seven year tenure at Wayfair that I've seen we just say yes to things without pushing any other project out."
>
> — Amanda Holland

## Restarting a database migration stalled for years

The database migration that started in the pilot is now moving 8-12x faster. Wayfair split it into four stages, and the first stage, 17 to 20 databases, was done or nearly done by mid-September, ahead of a plan to finish by the end of the year. Wayfair is also using Devin to break up the monolithic service behind supplier setup and to rebuild the platform that places product images on its listings.

## Spending efficiently

Fiona wants Wayfair’s teams to get more done while being thoughtful about cost, so they can put more time and resources into making the experience better for customers. Cognition built two controls Wayfair asked for within a few weeks: in-session coaching that suggests a cheaper approach, and budgets that follow an engineer across teams. Cognition also handles model routing, so engineers don't pick a model and Wayfair doesn't have to test every new one.

> "We want to make sure that our engineers don't have to make decisions on which model to use for their daily work, so Devin's model routing was a key capability for us."
>
> — Victor Andreoni Paseka

Victor treats efficiency as a skill.

> "We're not into token-maxxing... Being a good software engineer in this day and age is being able to use the tools in a way that also doesn't rack up the bills unnecessarily."
>
> — Victor Andreoni Paseka

Cognition runs training sessions with Wayfair's engineers on which tasks suit Devin and what planning to do before starting a session, so more tasks get done right the first time.

> "Cognition gave coaching that was specific to how we were using Devin at Wayfair. It was not just generic. It was how we use Devin to achieve our goals in our projects."
>
> — Victor Andreoni Paseka

## What's next

Fiona plans to put the extra capacity to work across the business, from how customers discover and buy products to the supply chain and the systems underneath. Victor's teams are building new features giving customers more choice over when and how their orders are delivered. Amanda's team plans to use automations so Devin can answer the triage questions it gets from the rest of the business, like why something happened and how to improve it.

> "The teams are truly flying and it's crazy how much we're getting done."
>
> — Victor Andreoni Paseka