Doug Clauson

As a principal product manager for Kiro, Doug Clauson has led product work on Kiro CLI, AWS’s command-line coding agent, and several of its core agent capabilities. At Amazon CodeCatalyst, he showed how an agent could turn a detailed issue into a pull request. At Kiro, he helped design the terminal interface where a developer could spot a wrong turn and redirect the agent before the change reached review.

From flight simulation to Amazon CodeCatalyst

Before joining AWS, Doug built flight simulation software and aircraft technology. He counts building X-wings for Walt Disney World among his aerospace projects. By 2023, he was a principal technical product manager in AWS DevOps Services and the product manager for Amazon CodeCatalyst. By 2026, he had spent five years managing developer tools at AWS.

At re:Invent 2023, Doug and Harry Mower, general manager of DevOps Services, explained how the Amazon Q preview in CodeCatalyst could propose an approach to an issue, write the code, open a pull request and watch for failures in automated checks. They recommended giving Q a detailed issue description. They also drew a distinction that mattered to anyone reviewing its work. Q was better at planning a change than implementing it, so even a sound proposal could lead to a faulty patch.

Steering Kiro during a change

Kiro CLI originally printed its conversation one line at a time. In the small terminal pane of a code editor, a developer could lose track of what the agent was doing. Doug and design lead Phill Shaffer worked on a full-screen terminal interface that kept the agent’s status visible. A developer who spotted Kiro working on the wrong file could interrupt it and point it to the right one with @. The team released the experimental interface behind the --tui flag so developers could try it on their own projects, give feedback and switch back.

When Doug and Romain Dura used Kiro workflows to build parts of Kiro itself, an agent could fill its context window writing a change before it reached review. They put the sequence of steps in a JSON or YAML recipe that a fresh session could follow. Implementation agents could work in separate Git worktrees without disturbing one another’s edits. If one needed a decision, it could ask the main session and get an answer while the remaining steps stayed in the recipe.

What a cheaper request still leaves to review

In November 2025, Doug described how Kiro’s Auto mode combined frontier and specialized models and used both caching and intent detection. For that update, Kiro reported that a typical request that had used one credit now used 0.79.

Doug’s upcoming AIE CODE 2026 session on Kiro Crew looks at the cost to the engineer shipping the change. When an agent turns a ten-line fix into a subsystem, one person still has to review the whole subsystem.