In the captured material, advisor strategy appears in the context of how teams deploy agents in production: the advisor pattern sits at the more conservative end of agent autonomy, where the system surfaces recommendations rather than executing changes on its own. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@1181] The channel's discussion of Anthropic's Claude platform — with Angela Jiang (head of product) and Katelyn Lesse (head of engineering) — frames why this tradeoff matters: the team argues agents should eventually be judged by just two parameters, an outcome and a budget, which presumes enough autonomy to act, not merely to advise. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@692] [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@113] The direction of travel described across the source is from synchronous prompting toward asynchronous delegation — you hand a task off, the agent spends significant time making calls and iterating, then reports back. An advisor strategy is the earlier, more constrained step on that path: the agent still does the analysis, but a person retains the decision. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@813] How the channel frames the tradeoff ## Advisor versus autonomous execution The material treats the advisor pattern as a risk-management posture rather than a technical architecture: what the agent does is unchanged (reasoning, tool calls, consulting data), but the approval step stays with a human. This matters because the real wall teams hit is not harness engineering but production infrastructure — reliability, transcript storage, and secure sandboxing — and an advisor posture bounds the blast radius while that infrastructure matures. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@1181] - Advisor: the agent recommends; a human decides and acts. - Autonomous: the agent executes within an agreed outcome and budget, then reports back. ### What the evidence leaves open The captured material does not fully resolve where to draw the line. The platform team's bet is that the outcome-and-budget framing eventually replaces the human approval step — but their own account shows eval results vary drastically with harness and model pairing, which is one reason teams keep an advisor checkpoint while infrastructure and evaluation practices catch up. [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@692] - An advisor strategy deploys an agent to recommend a course of action while a human keeps the decision and the execution. - It is the conservative end of the autonomy spectrum the channel contrasts with judging agents purely by an outcome and a budget. - Teams in the captured material favor this posture while production infrastructure — reliability, sandboxing, storage — is still the hard wall. Explore the moments above to hear how the platform team weighs autonomy against control in their own words.