Why You Must Be a Maximalist with AI

_{IN ONE SENTENCE}

Companies that test AI timidly miss the point: you need to push the system to do everything to discover what really works.

_{THE OBSERVATION}

Most companies adopt AI incrementally: a small POC here, an automated document summary there. Six months later, the POC was never scaled and the company concludes AI isn't ready yet. Ethan Mollick observes this pattern on repeat, and his diagnosis is clear: the problem isn't AI: it's ambition.

Having AI summarize documents is fine. But AI could already do that two years ago. The question Mollick asks is more fundamental: why not ask AI to do the actual work the document was meant to prepare, rather than just the intermediate step?

_{WHAT YOU NEED TO UNDERSTAND}

The permanent POC trap

Many companies remain stuck in the proof-of-concept phase. They deploy AI on a narrow use case, get decent results, then never manage to scale. The problem: they optimized for a case too small to generate real organizational change.

The maximalist approach in practice

Mollick recommends pushing AI to do everything: strategic analysis, content production, decision-making, prototyping, research. If it fails; excellent, you now have a benchmark to test future models. If it succeeds, you've discovered a source of value that the cautious approach would never have revealed. In 25 minutes with an advanced model and a terminal, Mollick can generate 25 ideas, evaluate them, simulate customer feedback, create a working prototype, and iterate on it.

The IT team bottleneck trap

A major obstacle: IT teams, often designated to lead AI deployment, naturally optimize for low latency and low cost. Yet for many strategic use cases, the right approach is the opposite: pay more for a smarter model. Mollick says it plainly; paying 15 cents for a brilliant strategic decision or a new molecule is a reasonable cost.

_{WHAT THIS CHANGES FOR YOU}

• Stop timid POCs: give AI the most ambitious tasks to see where it surprises you
• Every AI failure becomes a useful benchmark for future model versions
• Don't let IT alone decide which models to use: strategic use cases justify more expensive, smarter models
• Ask yourself the right question: can AI do the final work, not just the intermediate step?