Anthropic Lead Outlines 3 Strategies for Ambitious Prompting: Altitude, Depth, and Breadth

Felix Rieseberg, an engineering lead at Anthropic, published a tutorial in October 2026 addressing why many users fail to realize the full capabilities of AI models, often confining them to minor chores such as rewriting emails or populating templates. He attributes this limitation largely to low expectations and categorizes ambitious AI utilization into four dimensions: speed, altitude, depth, and breadth, focusing on the latter three for substantive operational impact. Increasing "altitude" entails shifting AI instructions from granular tasks to desired overarching outcomes. For example, rather than simply updating quarterly business review slides, users can prompt the model to redesign review workflows to detect performance issues earlier, or authorize it to autonomously fetch crash logs, trace git commits, and reproduce errors across dedicated operating environments. Deepening "depth" prevents models from merely adopting superficial expert jargon. Instead, it instructs AI to apply rigorous methodologies and tools used by top domain practitioners. This approach involves examining raw data, validating or challenging existing assumptions through databases and statistical models, and utilizing external verification systems to uncover blind spots. Broadening "breadth" reframes projects from static nouns into functional verbs, clarifying the core objective before convening multi-disciplinary perspectives—such as analysts, investigative editors, or forensic accountants—to stress-test plans via multi-agent collaboration. Rieseberg highlights that these approaches presuppose that AI systems possess environments capable of executing software independently and accessing external operational data via standards like MCP, noting that standard text chat boxes lack the depth needed for these outcomes.

Key takeaways

• Anthropic engineering lead Felix Rieseberg argues that user preconceptions limit AI to trivial tasks, urging a shift toward outcome-driven ambition.
• Raising 'altitude' shifts delegation from isolated tasks to end goals, restructuring workflows so models autonomously trace and verify fixes in isolated environments.
• Deepening 'depth' focuses on applying expert tools to raw data and challenging business assumptions, supported by external validation mechanisms.
• Expanding 'breadth' reformulates tasks as verbs to orchestrate cross-disciplinary multi-agent reviews, contingent on providing models with execution environments and data integrations.