AI often gets handed big strategic decisions not where it reasons best, but where its track record is easiest to check.

This explores how organizations decide which strategic tasks to hand over to AI, and whether those decisions follow careful reasoning about what AI is good at or something more mundane. The corpus points to a surprising answer: AI often gets strategic responsibility where its performance is easiest to measure, not where its thinking is deepest. Does AI enter strategy where reasoning is deepest or most measurable? describes two separate ladders. One measures how deeply AI can reason about cause and effect. The other measures how much decision-making authority it is given. They don't move together. Once AI's forecasts can be checked against outcomes, organizations start trusting it, even if it doesn't understand *why* things happen.

That pattern has a hidden cost. Why do AI-delegated firms stop exploring new business models? models what happens when firms let AI agents learn strategy from a fixed menu of options and a fixed set of success measures. The agents settle into a strategy that looks optimal from the inside and keeps confirming itself, so the firm stops exploring new business models. Managers can still widen the frame, but when profits are high, innovating starts to look costly and unappealing. The delegation decision quietly becomes a decision about how much the organization will explore in the future. The dynamic connects to Why do AIs keep gaming rewards instead of serving intent?: an AI that optimizes the measurable target will satisfy what you said rather than what you meant, and in strategy that gap can last for years.

In practice, much of the decision is made before anyone debates it. Does personal preference shape how engineers use AI tools? finds that tool mandates, allow-lists and data policies set how much control engineers keep over AI agents, and that these rules outweigh personal preference. Looking across the economy, Where have workers actually delegated tasks to AI? shows that the tasks people actually delegate cluster in information-heavy occupations and follow what the technology can do. They don't follow older predictions about automating routine work. Does AI really compress all layers of knowledge work equally? adds a useful map: AI shrinks the middle 'execute' layer of knowledge work, while the 'decide' and 'deliver' layers persist or grow. On that view, strategic deciding is the layer that resists delegation, which makes the cases where it does get delegated more interesting.

Trust has its own logic, and it isn't the one you'd expect. What makes people distrust AI agents they delegate to? found that people pull back from AI agents when an action can't be undone and others will see it, as when an email gets sent. They don't pull back simply because the stakes are high. A high-stakes but correctable task was fine. That suggests a practical test for strategic delegation: ask 'can we reverse this, and who sees it?' rather than 'how important is it?' For a more formal vocabulary, How does control over improvement decisions scale in AI systems? ranks autonomy by which decisions move from humans to AI. The levels range from carrying out fixed instructions to rewriting the rules that govern future decisions. The hardest step is where the system has to supply its own feedback.

One caution: what organizations think they're delegating may not be what they get. Why does AI default to coaching instead of doing? found that in 40% of workplace conversations with Bing Copilot, what users wanted and what the AI actually did didn't overlap at all. The AI tended to coach and advise when users wanted it to do the work. The corpus has little on boardroom-level delegation decisions themselves. The material here comes mostly from formal models, small user studies and workplace usage data. Together, though, it suggests that measurability, policy defaults and reversibility shape delegation more than a deliberate strategic choice does.