Three intelligences, three failure modes, one allocation decision your org chart hasn’t assigned.
THE GIST
Every piece of work now has three candidate intelligences — Analog, Automation, AI — each with its own failure mode.
Misallocation is expensive in every direction: 55% of AI-driven redundancy decisions are now admitted mistakes.
IBM routed the fork deliberately — 94% of routine HR automated, and total employment went up.
You’ve lived this meeting. The pilot demo dazzled. The roadmap says “AI-first.” And nobody in the room can say which work should actually move — or what breaks when it does.
Look at the org chart and try to find the person who owns that call. There’s an owner for security. An owner for data. An owner for compliance. There is no owner for the most consequential allocation decision of the era: for each piece of work, which intelligence leads?
There are three to route among, not two.
As work moves down the fork, the machine does less of the deciding — and the human carries more of the accountability.
Analog — human judgment. Only a person carries a stake in being right. But human judgment is inconsistent and doesn’t scale; stretch it across enough volume and it tires, varies, and misses. The status quo’s quiet leak.
Automation — static rules. Fast, consistent, tireless — and fixed. When reality moves, it keeps enforcing the old world. Point it at a broken process and you don’t fix the process. You industrialize it.
AI — reasoning on demand. Analysis, pattern, and generation at a scale no person matches. But it stakes nothing on being right, produces plausibility as readily as truth, and answers to no one.
None of the three fixes anything alone. The discipline is in the pairing — and in deciding, before the work begins, which one leads. That decision is the Intelligence Fork™.
Get it wrong in any direction and it’s expensive. In a 2025 Orgvue survey, 39 percent of business leaders said they’d made workers redundant because of AI. Fifty-five percent of those now admit the decisions were wrong — many rehired the same roles at higher salaries. That’s not a technology failure. That’s consequential, trust-bearing work sent down the machine path because it was technically automatable.
Now watch it done deliberately. IBM’s AskHR agent automates 94 percent of routine HR requests — 11.5 million interactions in one year. And IBM’s total employment went up, because the savings were re-routed into engineers and salespeople. The routine went to the machine. The judgment-heavy remainder stayed human. The freed capital went to work only humans can do. One fork, decided well.
TAKE IT TO THE FLOOR
List your ten most consequential decisions. For each one, ask: who decided whether a machine decides? If the answer is “whoever bought the software,” the fork is being decided by default — and defaults, at machine speed, compound.
See it at work: What we do and Results at BGR.
Sources: Orgvue 2025 survey of business leaders on AI-driven redundancies; IBM on its AskHR agent. Originally published at intelligencemanagement.com, where the full library of Intelligence Management™ essays lives.
