Most frameworks for decision-making answer the wrong question. They tell you when to escalate. That framing assumes escalation is the default and clarity is the exception, so leaders spend their time building better escalation paths instead of asking why so much is escalating in the first place.
The question this framework answers is the reverse one: what would have to be true for this decision not to be mine?
That single inversion changes the default direction. Instead of a team asking "should I check with the leader," the leader is asking "what evidence would justify this decision requiring me at all?" Most decisions, examined honestly, have no such evidence. They reach the leader because nobody ever made explicit that they didn't need to.
Level 1 — Autonomous. The team owns the decision and executes without informing the leader. This is the default for the large majority of operational decisions: implementation choices, day-to-day prioritization, anything where the person closest to the work has both the information and the standing to decide.
Level 2 — Consultative. The team decides and informs the leader afterward. The leader isn't part of the decision, but stays aware of the outcome. Appropriate when a decision is reversible and low-risk, but visibility still matters.
Level 3 — Shared. A co-decision. Both the team and the leader are required; neither owns it unilaterally. This tier exists for genuine boundary decisions, where the leader has context or authority the team doesn't, and the team has information the leader doesn't.
Level 4 — Escalated. The leader decides, with team input. The team's judgment shapes the decision, but the leader carries it. Reserved for decisions with real weight, where the leader's authority or accountability is the deciding factor.
Level 5 — Directive. The leader owns the decision entirely. No team input required. This is the rarest tier, reserved for decisions that are the leader's by definition, not by default.
Level 4 and Level 5 are meant to be rare. Using them requires a specific, documented reason, a corporate requirement, a defined compliance control, an explicit regulatory obligation, not habit, not comfort, not "that's how it's always been." Everything else defaults to Level 1 or Level 2.
Most organizations run this backwards. Decisions default upward because nobody ever built the case for keeping them at Level 1 or 2. The absence of an explicit decision-rights map isn't neutral. It's a structural bias toward escalation, and it's rational: a team with no stated authority will always choose the safer path of asking.
For any decision currently landing on your desk that shouldn't be, ask one question: can you name the specific regulation, control, or documented policy that requires this level of sign-off for this specific decision?
If you can name it, the decision is genuinely Level 4 or 5.
If you can't, one of two things is happening. Either it's an authority-information mismatch wearing compliance clothing, someone borrowed compliance language because it's unchallengeable, not because a real requirement exists. Or it's decision avoidance disguised as diligence, the question is answerable by someone on the team today, and the appeal to caution is providing cover for not answering it.
Regulated environments are the one place Level 4 and 5 usage should be higher than average, and even there, the discipline holds: name the actual control, not the general feeling that the environment is sensitive.
The same five-tier logic holds regardless of who or what is making the decision:
1. Applied to AI agents, the same structure governs what an agent is allowed to decide without review versus what requires a person to approve before it proceeds — Level 1 (decide and act, no reporting) through Level 5 (execute exactly as directed, no interpretation).
2. Applied to cross-functional boundaries (product and engineering, for example), the framework maps which decisions belong to one function outright, which require joint sign-off, and which sit above both, converting "whoever pushed hardest last time" into a designed interface instead of a relationship that keeps absorbing the same argument every quarter.
The framework doesn't change shape when the "team" is a group of engineers, a single AI agent, or two departments in permanent contact at a shared boundary. What changes is the specific content assigned to each level. The structure, and the discipline of keeping Level 4 and 5 rare and documented, stays constant.
This framework is part of LeadershipOS™, the operating system for calm, confident, high-performing teams. The full Stack, five layers, this framework, and how they interact, is in the book: TheLeadershipOSBook.com
I publish daily on structural leadership for executives and technical leaders. Subscribe: technicalleader.coach/daily-email
'
Want to see where your team's decision architecture is breaking down first? Start with the LeadershipOS™ Stability Scorecard: TheLeadershipOSBook.com/scorecard