What Kind Of Problem Even Is This?

There's a specific kind of delegation failure that doesn't respond to the usual fixes. You weren't unclear about the steps. You weren't vague about success criteria. You got exactly what you asked for, executed competently, and it was somehow still useless, so the problem must be somewhere else entirely.

Task and Specification

Most advice about delegation operates on two axes. Task delegation: tell them what to do. Specification delegation: tell them what good looks like. These map onto familiar dichotomies—process versus outcome, imperative versus declarative. For problems with clearly articulated procedures or success criteria, they're exactly right. Didn't get what you wanted? Be clearer about the steps, or be clearer about the rubric.

But some problems resist delegation at either level.

The Research Request

Consider: "Find out what we should know about X before the meeting."

Task delegation would be: search these sources, summarize these topics, organize by theme. Specification delegation would be: cover risks, opportunities, competitive landscape, and timeline; keep it under two pages.

You can make either perfectly precise. And you can get back a document that hits every mark—correct sources, right length, all topics addressed—that completely misses the point. Because what you actually needed to ask was why you are asking in the first place. What would change your mind. What would be surprising versus confirmatory. What's signal for this decision versus noise that happens to be nearby.

That's not task and it's not specification. It's something else.

Orientation

Call it orientation. Where task and specification operate at the instance level—this procedure, this rubric—orientation operates one level up. You're not delegating the work; you're delegating the question of what counts as appropriate work. What kind of problem is this? What kind of thinking applies? What makes something success for this kind of situation?

Think of a coach who can't step onto the field. You can't call every play. But you can provide the strategic frame that guides tactical decisions in situations you didn't anticipate. "Tire them out in the first half" isn't a play and it isn't a score target—it's an approach that tells your team how to evaluate options when you're not there to make the call. Delegating orientation aims to transfer that capacity: enough context about why that someone can recognize what matters, not just execute what you specified.

The Same Request, With Orientation

Return to the research request, but framed differently:

"We're 80% committed to this acquisition. At this point I'm not looking for comprehensive background—I'm looking for reasons to pump the brakes. What would make us reconsider? Steelman the argument for pulling out. If you find nothing surprising, that's genuinely useful signal. But don't waste time on stuff that defends the current plan."

This doesn't specify sources or format. It doesn't provide a rubric for completeness. But it transfers something more important: the cognitive posture. The recipient now knows what to ignore (confirmatory information, however accurate), what to flag (anything that could change the decision), and what "done" feels like (a stance on whether to proceed, not a comprehensive summary).

The same facts get filtered differently. A competitor's recent hire might be noise in a background document but signal in a stress test. The orientation makes the task and specification derivable in context—because now you know what you're actually trying to learn.

What to Notice

Orientation shows up in language that signals stance rather than steps or criteria:

Cognitive mode: "We're stress-testing, not polishing." "I'm trying to poke holes." "This is a sanity check, not a deep dive."

Phase: "We're early—I need the full option space before narrowing." "We're late—I need to know what could still go wrong."

Purpose: "I need ammunition for a skeptical audience." "I want to know what a thoughtful critic would say."

None of these are tasks. None are specifications. But each represent useful language in a prompt that tells the AI what success feels like for this kind of work.

Defaults

Here's what makes orientation easy to miss: you always get one, whether you provide it or not.

A longtime colleague brings rich defaults. They've sat in the meetings, heard the hallway conversations, watched how you react to certain kinds of news. When you ask for research on an acquisition target, they're drawing on everything they know about your risk tolerance, your past decisions, what keeps you up at night. Your prompt is a thin slice; their context is thick.

AI sees only the prompt. Everything else—the meetings, the history, the organizational dynamics—is largely invisible. What fills that gap is training: strategies that work broadly across many problems for many users. Be thorough. Consider alternatives. Cover the angles.

This is what makes AI good at coding and math—problems where systematic reasoning toward a verifiable answer is exactly right. But that same training creates a default orientation: when given a problem, solve it. Not ask "what kind of problem is this?" or "is this the right problem?" but return "here's the comprehensive answer to what you asked."

The colleague might push back: "Wait, is this actually what you need?" The AI assumes you've already figured that out.

Recognizing the Gap

Naming this helps us solve the issue. When output is disappointing despite clear instructions, you have question to ask: did I give the AI a stance to orient it, or did i just give it a task and specification and hope it took the right stance?

Think of it like working with a sharp junior. You wouldn't hand them a research task and expect useful output without context—not because they lack capability, but because thinking work requires understanding what kind of thinking you're doing. You'd tell them: "We're stress-testing a decision we've mostly made" or "We're early—I need to see the full space before we narrow." Not because they need to learn for next time, but because they can't help with this problem without knowing your stance.