You are already delegating at multiple levels. You just haven't named them.
Right now, you delegate differently to different people, for different tasks, in different contexts. You give your most trusted colleague a vague outcome and trust them to figure it out. You give the new hire a step-by-step checklist. You ask your partner for input on a decision you've already mostly made.
None of that is wrong. What's wrong is that you're doing it unconsciously — and the person on the receiving end has to guess which level you intended. That guessing game is the source of most delegation failures: not incompetence, not bad faith, but a mismatch between the autonomy you meant to grant and the autonomy the other party assumed they had.
The fix is embarrassingly simple. Stop treating delegation as a binary (do it yourself or hand it off) and start treating it as a ladder with distinct, nameable rungs.
The spectrum that every delegation model discovers
The idea that delegation exists on a continuum — not as an on/off switch — has been independently discovered by researchers and practitioners across decades.
Tannenbaum and Schmidt (1958) published the first formal model in the Harvard Business Review. Their leadership continuum identified seven positions between "manager makes decision and announces it" and "manager permits subordinates to function within defined limits." The key insight was not the specific positions but the underlying principle: authority and autonomy exist in inverse relationship along a single axis. Moving right gives the team more freedom. Moving left gives the leader more control. Neither extreme is inherently correct — the question is always which position fits the situation.
Hersey and Blanchard (1969) built on this with Situational Leadership Theory, arguing that the correct level of delegation depends on the follower's readiness — a combination of ability (can they do it?) and willingness (will they do it?). Their four styles — Telling, Selling, Participating, Delegating — map to four readiness levels. A person who is both able and willing gets full delegation. A person who is unable and unwilling gets direct instruction. The model has been widely adopted despite limited empirical validation — Thompson and Vecchio (2009) found that the 2007 revised theory was actually a poorer predictor of subordinate performance than the original version. The research is mixed, but the core principle holds: delegation level should match the delegate's current capacity for the specific task.
Jurgen Appelo (2011) produced the most actionable version in Management 3.0. His seven levels of delegation give each rung a single verb:
- Tell — You decide. You may explain your reasoning, but the decision is yours.
- Sell — You decide, but you actively work to get buy-in. You want them to feel involved, not just informed.
- Consult — You ask for input first and take it seriously, but you make the final call.
- Agree — You discuss as equals and reach consensus. No one has unilateral authority.
- Advise — You offer your opinion and hope they listen, but the decision is theirs.
- Inquire — They decide first. Afterward, you ask them to explain their reasoning.
- Delegate — They decide. You don't need to know the details.
Appelo's contribution was making the ladder operational. His Delegation Poker game has teams play numbered cards (1-7) simultaneously for each decision area, forcing the mismatch into the open. When one person plays a 2 (Sell) and another plays a 6 (Inquire) for the same decision, the gap becomes visible — and negotiable — instead of lurking as an assumption that detonates later.
Three models. Six decades. The same discovery: delegation is a spectrum, the rungs are distinct, and most failures happen because people skip the step of explicitly naming which rung they're on.
Why the middle rungs matter most
Binary thinkers default to the extremes. Level 1 (Tell) is micromanagement. Level 7 (Delegate) is abdication. Both feel decisive. Both are usually wrong.
The real work of delegation happens in Levels 2 through 6, where authority is shared in specific, negotiated ways. Consider how radically different each middle rung is:
Sell vs. Consult: In Sell, you've already decided — you're just building commitment. In Consult, you genuinely might change your mind based on input. The delegate's experience of these two is completely different. If you say "I'd love your thoughts" but have already decided, you're performing Consult while operating at Sell. People detect this within one or two cycles, and it destroys trust faster than simply telling them the decision.
Agree vs. Advise: In Agree, neither party can override the other — you must reach consensus. In Advise, you've ceded decision authority and are offering counsel. The difference determines who has veto power. Confusing the two means someone thinks they can block a decision when they can only influence it, or thinks they own a decision that actually requires your agreement.
Inquire vs. Delegate: Both give the delegate decision authority. But Inquire retains a feedback loop — you want to understand their reasoning after the fact. Delegate cuts even that cord. The difference matters because Inquire builds your model of how they think (which lets you calibrate future delegation), while Delegate is a statement that you trust both the outcome and the process and don't need visibility into either.
Each rung is a different contract. Using the wrong one doesn't just produce the wrong outcome — it produces the wrong relationship.
Matching the rung to the situation
The ladder is not a maturity model where higher is always better. Level 7 for a task the delegate cannot handle is recklessness. Level 1 for a task they've mastered ten times is disrespect. The right rung depends on three variables:
Task complexity and reversibility. Low-stakes, reversible decisions can tolerate higher autonomy levels. The cost of a wrong call is low and the learning value of making the call is high. High-stakes, irreversible decisions — merging a company, terminating someone, shipping a security-critical change — may warrant lower levels even with highly capable delegates. The question is not "do I trust this person?" but "can we afford the cost of this decision going wrong while the delegate is learning?"
Delegate competence for this specific task. Not general competence — task-specific competence. Your best engineer might need Level 2 (Sell) for a people-management decision and Level 7 (Delegate) for an architecture choice. Hersey and Blanchard got this right even if their broader model lacks empirical rigor: readiness is task-specific, not person-specific. Calibrating the rung per-task rather than per-person is what separates effective delegation from lazy categorization.
Trust trajectory. Deci and Ryan's Self-Determination Theory (1985, 2000) established that autonomy is a basic psychological need — people function better and are more intrinsically motivated when they experience genuine choice in their actions. This means the ladder should generally trend upward over time. Not because higher is morally better, but because competence grows with practice, and autonomy fuels the motivation that accelerates that growth. A delegation relationship that stays at Level 2 for a year is either misjudging the delegate's growth or failing to develop it.
The practical protocol: start at the rung the situation demands, then deliberately move up one level when the delegate demonstrates readiness. Each successful step up is evidence that justifies the next. Each failure is signal to step back one rung — not collapse to Level 1.
The AI delegation ladder: same logic, higher stakes
Every interaction with an AI agent is a delegation decision. And the same seven-level logic applies — but most people operate at the extremes. They either dictate every step to the model (Level 1: Tell) or hand off entire workflows with minimal oversight (Level 7: Delegate). The middle rungs are where productive human-AI collaboration actually lives.
Anthropic's research on Claude Code users (2025) reveals the ladder in action. New users (under 50 sessions) employ full auto-approve — effectively Level 7 — about 20% of the time. By 750 sessions, that rises to over 40%. But here's the nuance: experienced users also interrupt more often (9% of turns vs. 5% for newcomers). They've moved up the ladder not by removing oversight but by shifting from pre-approval (Level 3: Consult) to post-hoc monitoring (Level 6: Inquire). They let the agent act, then review selectively. That's a sophisticated delegation stance that most people skip when thinking about AI.
The SAE levels of driving automation (J3016) formalize this same progression for autonomous vehicles across six levels from no automation (Level 0) to full automation (Level 5). The critical boundary is between Level 2 and Level 3: below it, the human monitors continuously. Above it, the system monitors itself and requests human intervention only when needed. That boundary — who is responsible for monitoring — is the same question you're answering every time you set an AI agent's autonomy level.
For your own AI delegation practice, the Appelo ladder translates directly:
- Tell: You dictate the exact prompt, format, and steps. The AI is a typewriter.
- Consult: You describe the problem and ask the AI for options. You choose.
- Agree: You iterate with the AI, co-developing a solution through conversation.
- Advise: The AI generates a plan. You flag concerns but defer to its judgment.
- Delegate: The AI executes a multi-step workflow autonomously. You review outcomes, not process.
The right level depends on the same three variables: task reversibility, your calibrated understanding of the model's capability for this specific task, and the trajectory of trust you're building through repeated interactions.
The delegation ladder as epistemic infrastructure
The ladder is not a management technique. It is a piece of cognitive infrastructure — a named, graduated scale that replaces a vague feeling ("I should probably give them more autonomy") with a precise operating position ("I'm at Level 3 for this task and I'll move to Level 5 when they've demonstrated X").
This matters because unnamed levels cannot be negotiated, calibrated, or deliberately adjusted. When you say "I want you to handle this," the delegate doesn't know if that's Level 5 (Advise — they decide but you want input), Level 6 (Inquire — they decide and explain after), or Level 7 (Delegate — they decide and you don't need to know). The ambiguity is not a minor friction. It is the mechanism by which most delegation relationships fail.
Name the rung. Every time. For every task. For every delegate — human or AI.
The person who delegates without naming the level is hoping their intent is obvious. The person who names the level is building a system where delegation improves by design rather than by luck.
That system — the capacity to delegate at progressively higher levels across more domains — is what the next lesson addresses: building delegation capacity.