GPT-5: It Just Does Stuff

Ethan Mollick explores how GPT-5 demonstrates a fundamental shift in AI capabilities: the move from passive response systems to active agents that can initiate and complete work independently. Rather than waiting for user input at each step, newer AI systems can evaluate situations, make decisions, and execute tasks with minimal human intervention—a capability that defines true agentic workflows.

This represents more than a performance upgrade. Mollick emphasizes that the meaningful change is behavioral: AI moving from 'answering questions' to 'doing things.' For professionals, this means rethinking how AI integrates into daily work. Instead of using AI as a research or brainstorming tool, organizations can direct AI agents to handle complete projects—from planning through execution.

The implications span across industries. In professional services, agents could manage client communications and project tracking. In healthcare, they might handle administrative workflows. In education, they could personalize learning experiences autonomously. The catch is that autonomous action requires careful setup and oversight—defining clear boundaries, goals, and failure modes becomes critical when AI operates without constant human approval.

Mollick's core insight challenges how we think about 'putting AI in charge.' It's not about replacing human judgment but about delegating well-defined operational tasks while humans focus on strategy, oversight, and decisions requiring judgment or ethical consideration.