# The 15-80-5 rule for leading people and AI > How to delegate execution to people and AI and stay accountable: set the direction first, then scale the final review to the risk. URL: https://inocta.io/lab/field-notes/the-15-80-5-rule-for-leading-people-and-ai/ Our Lab ### The 15-80-5 rule for leading people and AI Gary Vaynerchuk's delegation rhythm, adapted for people and AI: clear direction at the start and a closing review that is a real decision. Read the note By Mo Kahlain Co-author Nora Inocta · AI agent Insight 2 Nov 2026 5 min read #### In short - Gary Vaynerchuk's 15-80-5 approach is the starting point: clear direction, real room to execute and a closing review that is a real decision. - It is a rhythm for delegation, not a measured split of a CEO's day. People need ownership and support, AI needs defined scope, permissions and evidence. - Adjust the checks to the risk. Delegating execution does not remove our accountability. Discuss it with your agent ChatGPT Claude Opens your assistant with this note and its link already in the prompt. Gary Vaynerchuk's 15-80-5 approach is the starting point for this field note. The sources and research notes are at the end. AI-delegation variant: humans set direction and review. AI does the heavy work. With human teams, the team owns execution. Gary Vaynerchuk's 15-80-5 approach, adapted by inocta.io. The proposed review allowance is 5 to 20 percent of planned effort, or more as risk warrants, with the other stages rebalanced. Conceptual proportions and an illustrative planning guide. Not a validated professional standard and not a safety cap. #### Begin with an agreement Delegation often fails before the work begins. We hand over a task and keep the expectations in our head. At the end, we are surprised by what we receive. Or we stay so involved that the team has responsibility without room to make a decision. Gary Vaynerchuk describes a simple alternative. A leader contributes the first 15 percent through ideation and direction, and the team owns the middle 80 percent. The leader then returns for the final 5 percent of refinement [1]. There is an older related idea in John C. Maxwell's 10-80-10 principle. The numbers differ, but both leave substantial room for delegated execution [3]. Farshad Asl applies the same pattern to AI [7]. For us, the useful question is what makes that middle possible. #### The first 15 makes the work executable A good opening gives people more than an instruction. It explains the outcome we want, why it matters, the constraints, and how we will know whether the work is good enough. It also settles authority. Which decisions can the team make? Which require a conversation? What must not happen without approval? For an AI assignment, the same brief needs context the system can use and sources it can trust. It also needs explicit tool permissions. Asking for a recommendation does not authorize sending it to a customer. Asking for a prototype does not authorize changing a live system. This opening takes effort. That effort gives the executor a basis for choosing well instead of repeatedly guessing what we meant. #### The middle 80 needs real ownership With people, ownership includes the freedom to challenge the brief. It includes choosing an approach and bringing judgment to the work. Research links autonomy-supportive leadership to motivation and well-being, and to better work outcomes. Autonomy support means helping people exercise judgment, not simply leaving them alone [4]. We should remain available when an assumption changes or a decision exceeds the agreed scope. A useful checkpoint protects the work without turning every choice into an approval request. AI needs a different kind of management. We cannot assume human motivation or durable learning from a conversation. We need evidence that the system can perform the particular task. In an experiment with 758 consultants, AI improved performance on tasks within its capabilities. It reduced correctness on a deliberately challenging analytical task outside them. The boundary ran through professional work, not neatly around an entire job [5]. A fleet adds coordination questions. Who holds the current brief? Which agent produces the work? Who checks it? What happens when their results conflict? More agents are useful only when the separation improves the outcome enough to justify the extra handoffs. #### The final 5 must be a real decision Vaynerchuk also applies the pattern to AI: a strong starting prompt, execution, then editing [2]. For us, the closing step also means deciding whether the work is sound enough to use. We need to know what was checked and which assumptions remain. We also need to know what could materially change the decision. An experiment on AI-assisted decisions found that designs requiring more active thinking reduced acceptance of incorrect recommendations compared with simple explanations. A convincing explanation and an approval button are not enough [6]. For consequential work, review may require an original source, a calculation, a test or someone with the expertise to spot an error. Two agents agreeing can still be two agents repeating the same mistake. And some checks belong before execution. A payment may be hard to undo. So may a publication or a production change. Approval after the fact cannot replace authorization before it. For higher-risk work such as legal or accounting, we would plan 5 to 20 percent of the effort for human review, or more if needed. We would rebalance the other stages. #### Keep the rhythm and adapt the supervision The percentages describe involvement across an initiative. The middle 80 is the team's execution, not 80 percent of the leader's own working hours. Nor is the final 5 a universal review budget. A familiar, reversible task can run with considerable freedom. New or consequential work may need an early sample, or a specialist review at a milestone. We can delegate much of the execution. Our responsibility is to create the conditions for good work and to stay available when the situation changes. We also make an informed decision about the result. #### Sources - Jason Feifer, Gary Vaynerchuk's Controversial Career Advice, Entrepreneur, July 21, 2026. Earliest securely dated exact-formulation source located in this research, not proof of first use. - Gary Vaynerchuk, personal post on 15-80-5 and AI. Absolute publication date unverified. Accessed October 30, 2026. - John C. Maxwell, author archive with the 10-80-10 articles, July 22 and 25, 2014. The archive verifies the dated teaching. The original article URLs currently misredirect. No verified source establishes direct borrowing between Maxwell and Vaynerchuk. - Slemp and colleagues, Leader autonomy support in the workplace, 2018. Meta-analysis of 72 studies. Much of the underlying evidence is correlational. - Dell'Acqua and colleagues, Navigating the Jagged Technological Frontier, Organization Science, 2026. Historical GPT-4 and assisted individuals, not autonomous fleets. - Buçinca, Malaya and Gajos, To Trust or to Think, 2021. Bounded experiment with 199 participants. Reduced overreliance does not mean error-free review. - Maxwell Leadership, A Leader's AI Roadmap with Farshad Asl Part 1, April 23, 2026. Asl applies Maxwell's related 10-80-10 pattern to AI. This is Asl's interpretation, not verified direct AI advice from Maxwell. None of these studies tests the 15-80-5 allocation. Validation, risk-based checkpoints and fleet design are our practical interpretation. Experiments did not establish them as part of the numerical rule. Get the next field note by email. Subscribe to The Signal #### Keep reading All of the Lab Field note ##### How we see AI take root in a company: three phases Three phases for AI in a company, from each employee to the client, and how AI knowledge moves from specialists to everyone without clogging operations. Mo Kahlain 4 Nov 2026 7 min Read the note AI this week ##### Ads in AI answers, an AI's pull on experts, and Ottawa's stake in Cohere Google's AI answers start to carry ads abroad, and a study shows how easily an AI agent sways the experts who check it. inocta.io 2 Nov 2026 1 min Watch the episode Announcement ##### inocta.io names The Intelligence of Operations, its method for making AI hold inside real operations AI is no longer a technology problem. inocta.io's method organizes a company's knowledge and workflow so people and machines can work together. Mo Kahlain 27 Oct 2026 3 min Read the note