What CEOs Need to Know About Artificial General Intelligence
AGI's timing is genuinely contested, so CEOs should not bet on its arrival but align operating models to a persistent 'jagged frontier' of uneven AI capability, where advantage comes from how AI augments people.
Summary
BCG argues that nobody can prepare for true AGI, which would erase humans' cognitive edge, so CEOs should plan for AI's uneven jagged frontier persisting. In a 2025 AAAI survey 76% of researchers judged scaling current approaches unlikely to reach AGI, and bottlenecks include data exhaustion, limits of verifiable-reward reinforcement learning, no on-the-job learning and compute ceilings. Government gating of frontier models could also cap impact. A Pascal's wager logic favors preparing for jaggedness, with advantage in token-based competition built on change management and skills.
Editor's opinion
This is a rigorously argued piece rather than an assertion of a framework: it builds a genuine technical case -- four distinct bottlenecks, a live scaling-versus-architecture debate, a named survey statistic -- for why AI’s jagged frontier may be structural rather than a temporary artifact of insufficient scale. It’s also unusual in treating regulatory intervention as a strategic variable in its own right, backed by dated, named, verifiable examples rather than a general gesture at “regulation.” The Pascal’s Wager framing is a clean, honest way to make a decision-under-uncertainty argument explicit rather than hiding it inside a confident prediction -- and the closing “token-based competition” idea gives CEOs an actual lever (augmenting people, not just deploying models) rather than leaving the piece as pure diagnosis.
Key concepts
Jagged frontier · Pascal's wager · Token-based competition · Reinforcement learning with verifiable rewards