Verification Gap — why AI structurally cannot replace strategy

AI can write code at senior level. But it can't develop strategy. That's not a coincidence — and based on the current state of technology, it's not a problem that can simply be solved with more compute.

Andrej Karpathy — OpenAI co-founder, former Tesla AI chief — recently said this openly: AI models learn through reinforcement learning, through automatically verifiable feedback. Code compiles — or it doesn't. Math checks out — or it doesn't. Financial modeling meets the constraints — or it doesn't.

A good strategy? Can't be automatically evaluated. A negotiation? Neither. This is called the Verification Gap — and it remains unsolved in current research.

OpenAI tried it internally as the "Universal Verifier": an AI that could evaluate any answer. It remained vaporware. Recently, the RL lead responsible left the company.

For companies, this creates a clear distinction:

Verifiable Workflows — code, math, financial modeling, document review: AI takes over, scales, optimizes. Fuzzy Domains — strategy, negotiation, culture work: Humans remain indispensable by current standards, AI can support.

This distinction is the strongest argument against the "AI replaces everything" narrative in board conversations. And the most honest basis for AI business cases.

The Verification Gap isn't a flaw. It shows where AI investments pay off — and where they don't.