Zenter Framework
The Three Gates Framework
A decision architecture for AI in real work.
Most AI rollouts start with the wrong question. They ask where AI can be used, then chase that list across the organisation. The better question is narrower and harder: for this specific task, should AI do it, and what do we deliberately keep human?
The Three Gates is a sequence for answering that, one task at a time. Three questions, asked in order. A task passes only if it clears all three. Most of the failures in AI adoption come from skipping a gate, usually the last one.
Gate One
Can AI do this?
A capability test, and the one most organisations get wrong because they answer it from a demo. A demo shows what the tool can do convincingly. The gate asks what it can do reliably, at your volume, on your real inputs, without a specialist hovering over every output.
The honest version separates two things the hype collapses together: tasks AI performs well, and tasks AI performs plausibly but unreliably. The second category is the dangerous one. It looks like a pass and behaves like a liability.
Gate Two
What's at stake if it gets it wrong?
A task AI can do is not a task AI should do if the cost of an error is high and the error is hard to catch. This gate weighs three things: the cost of being wrong, whether the mistake is reversible, and whether anyone would notice before it does damage.
Low stakes and reversible, let it run. High stakes, hard to reverse, easy to miss, that is exactly where a human stays in the loop, regardless of how capable the tool looked at Gate One.
Gate Three
What happens to our people?
This is the gate that gets skipped, and it is the one that decides whether adoption builds capability or quietly erodes it. The usual version treats people as a displacement question to manage after the fact. The strategic version asks something sharper: what is the deliberate value of keeping a human here?
In a market where everyone has access to the same models, the work you choose to keep human becomes the thing that distinguishes you. Judgment, accountability, the relationship, the sentence only a person can write. This gate is an investment decision, not an HR afterthought. Some humans you keep in the loop because the task is risky. Others you keep because their presence is the point.
How to use it
Run the gates in order, task by task, not system by system. The interesting tasks are rarely the ones that fail at Gate One. They are the ones that clear capability and stakes, then meet a Gate Three answer that says keep a human, in a redesigned role, doing the part that actually matters.
That is the output the framework is built to produce. Not a list of what to automate. A clear-eyed map of what to redesign, and what to protect.
