Why Executives Love AI and Engineers Don't — The Determinism Divide Explained

If you've been in any tech company's Slack workspace over the past 18 months, you've seen the pattern: executives pushing AI mandates, individual contributors pushing back. The debate plays out in engineering forums, internal channels, and Hacker News threads with striking consistency. A new essay by software engineer John J. Wang offers what may be the most cogent structural explanation for why.

The Core Thesis: Determinism Tolerance

Wang's argument centers on a fundamental difference in how executives and individual contributors (ICs) are evaluated and trained to think:

Executives operate in non-deterministic systems by design. Managing organizations means working with incomplete information, misaligned incentives, and emergent behaviors that no model perfectly predicts. A manager's job is to build a worldview and align utility functions across a chaotic system — accepting that specific outcomes are unpredictable even when overall system dynamics are understood. AI's non-determinism is familiar territory.

ICs are evaluated on deterministic execution. A software engineer is responsible for code that either works or doesn't. Tests pass or fail. Specifications are met or not. The hallucination rate of an LLM — a fundamental probabilistic property of the technology — is not a known-and-acceptable variance. It's a defect. The "it usually works" quality guarantee that satisfies an executive is a production incident waiting to happen for the engineer shipping the feature.

Why This Matters for AI Adoption Strategy

The implication is that AI adoption mandates imposed from above will structurally fail to account for how engineers actually experience AI tools. An executive who uses Claude or GPT-4o to draft strategy memos and synthesize research sees an impressively capable non-deterministic assistant. An engineer who uses the same tool to write infrastructure code encounters a system that confidently generates plausible-but-wrong code that silently breaks in production.

These are not the same product, experienced through the same lens. They're two different relationships with uncertainty, and the gap between them explains more about enterprise AI adoption friction than any particular product limitation.

The Management Implication

For AI product teams and executives designing adoption programs, Wang's framework suggests a design challenge: build workflows that abstract over AI's non-determinism in ways that matter to ICs, not just to managers. The engineers who need convincing aren't asking "is this impressive?" They're asking "can I stake my engineering reputation on this?"

That's a harder bar — and a more honest one.