Corey Quinn

Corey Quinn is the co-founder and Chief Cloud Economist at Duckbill and the creator of Last Week in AWS. He built a consulting practice around a question an AWS bill rarely answers on its own: which engineering decisions made it cost that much? His writing applies the same scrutiny to new services and pricing plans, showing where the fine print changes what a team should buy or build.

From automating servers to explaining AWS bills

Corey started as a Unix systems administrator working on large email systems. He moved into data center automation and contributed early code to SaltStack. As teams began replacing servers with fresh copies of a known configuration instead of fixing the machines in place, he reconsidered his specialty. He founded Quinn Advisory Group around a problem he had already solved several times: bringing an unexpectedly large AWS bill under control.

A new AWS service or pricing change could alter the economics of a client's system, so Corey automated the collection of updates. That gave him the raw material for Last Week in AWS, which he launched in 2017. He picked announcements worth a closer look, paired them with accounts of how engineers actually used the services, and made jokes about what the announcements left out. He also hosts Screaming in the Cloud, where he interviews cloud practitioners about how and why they build and operate their systems.

In 2019, Corey and Mike Julian combined their consulting businesses to form The Duckbill Group. They had spent years hiring each other for projects. Mike became CEO and Corey took the cloud economist role. After the firm abandoned an earlier software idea, Corey challenged Mike to build the kind of cost management product they kept telling vendors was missing. The team launched Skyway in 2025 and shortened the company name to Duckbill. Skyway's first module put contract terms beside actual usage so customers could see how quickly they were spending against a commitment and check whether their negotiated discounts were showing up on the bill.

What the advertised limits leave out

In his analysis of CloudFront's flat-rate plans, Corey worked out the average response size at which a customer would use up the plan's request and data transfer allowances at the same time. A software download can move a lot of data in one request. An API that sends small responses can exhaust its request allowance with most of its data transfer allowance still unused. The same plan can therefore be a bargain for one workload and a poor fit for the other.

What AI systems depend on

Corey argues that trace fields are an interface for agents. An error chart can show when failures spiked, but an agent needs records of individual requests to find out what the failures had in common. That makes a field rename consequential: a prompt that still asks for the old field can quietly miss the evidence it was written to find.

In Artificial Confidence, his newsletter on the business of AI, Corey looks at a dependency that starts before a system ships: whether you rent access to a hosted model or can run a copy yourself. A downloadable model may be less capable, but the provider cannot take your copy offline. That tradeoff matters when a team is choosing what to build around, even if the hosted model performs better today.