Where the term comes from in this material ## How verifiable outcome is used here In this channel's conversation with Angela Jiang (head of product, Claude platform) and Katelyn Lesse (head of engineering, Claude platform), the term appears in the context of Claude Managed Agents — Claude in a cloud computer with memory, tools, and infrastructure. The team's position is that agents should eventually be judged by just two parameters: an outcome and a budget. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@113] The emphasis on verifiable matters because it changes how the work is framed: instead of measuring how many steps the agent took or which tools it touched, you specify the end state you care about — the task is done when that end state demonstrably holds. This pairs with the shift from synchronous prompting to asynchronous delegation: Managed Agents launched with outcome-based tasks where the agent works asynchronously and returns only when done. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@113] The budget side supplies the other half of the contract — how much the agent may spend reaching the outcome — but the outcome is the thing you actually check. - A verifiable outcome is a task defined by an end state you can check as achieved, rather than by the process used to get there. - In this material, it is one of the two parameters — outcome and budget — by which the Claude platform team argues agents should be judged. - It underpins outcome-based tasks in Claude Managed Agents, where the agent works asynchronously and reports back only when done. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@113]