AI/Human Split
The allocation rule inside Woodring's Loop: give the volume work to the machine and keep the judgment work for the people whose names are on the decision.
What AI/Human Split is.
AI/Human Split. The AI/Human Split is the discipline of assigning AI to high-volume tasks — processing, pattern surfacing, draft generation, consequence modeling — while leaders keep the judgment tasks that require accountability. The line is not "trust but verify"; it is that no model owns anything with consequences unless a named person stands behind it.
What it is
The AI/Human Split defines who does what across the planning function. AI does the volume — reading thousands of filings, clustering signals, running sensitivity analysis, watching leading indicators. Humans do the judgment — sizing stakes, culling what does not matter, holding accountability, and moving a leadership team to a shared commitment. It exists to defend against two failure modes: Oracle worship, where executives rubber-stamp confident model outputs, and Oracle dismissal, where they ban the tools and get out-run by competitors running the Loop at five times their throughput.
How to apply it across the four phases
In Scan, AI earns its keep most — volume triage, semantic clustering, and anomaly detection on data no human team can process at speed. In Story, AI drafts candidate scenarios and red-teams the set to find the missing one, but never authors the scenario set. In Stake, AI models consequences across dozens of variables in minutes, then stops — the bet is the leader's. In Steer, AI watches the signals each scenario implied and flags divergence continuously, while the decision to adjust, hold, or reverse stays human. In every phase the pattern holds: AI does the volume work that has been the bottleneck, humans do the judgment work that has been the point.
The principle
The AI/Human Split is the operational form of Principle 4 — Judgment over machine. The Loop is engineered to be human-led and machine-accelerated; reverse the allocation and the discipline collapses into automation of the decision function, which is what the discipline is built against. Every AI-touched output carries a human reviewer's name, and when the model is wrong — it will be — the person who used it owns the consequence. Volume to the machine, judgment to the leaders.
Common questions.
- Isn't the AI/Human Split just "trust but verify"?
- No. Trust but verify still lets the model own the work between checks. The AI/Human Split is stricter: no model owns anything with consequences unless a named person stands behind it. Verification is the floor, not the rule.
- What can AI never be delegated in the Loop?
- Four things resist delegation: judgment under ambiguity, accountability, knowing what does not matter, and organizational politics. Concretely, never delegate the composition of the scenario set, the sizing of any stake, or the Steer call to adjust, hold, or reverse a commitment.
- How do you keep the AI/Human Split from eroding over time?
- Tag every AI-touched output with the human reviewer's name and run The Judgment Audit — a two-hour exercise that maps each Loop activity to a named accountability owner — at install and every six months. The accountability chain dissolves quietly when teams see throughput climb and stop tagging.
Source: Chapter 10 — AI as Planning Partner · Predictive Planning (Colloquial Media, 2026)
More of the vocabulary.
- Woodring's Loop
The four-phase continuous cycle at the core of predictive planning — Scan, Story, Stake, Steer — that turns weak signals into strategic decisions.
- Cycle Audit
A diagnostic that measures what your annual planning cycle actually costs — in senior attention, budget, and frozen strategic conversation — before you decide to retire it.
- Stake Sizing Matrix
A decision tool in Woodring's Loop that sizes any strategic commitment against two axes — the team's conviction in the scenario and the cost of being wrong.
- Convergence Review
A monthly leadership meeting that classifies every active scenario as converging, holding, or collapsing — and moves the stakes attached to each accordingly.