The Power to Shape AI

What It Covers

Professor Ethan Mollick's essay "The Shape of the Thing" traces AI's evolution from 2023 chat-based cointelligence to 2025 agentic systems, arguing that despite widespread feelings of helplessness around job disruption and market instability, individuals and organizations retain meaningful power to shape AI's trajectory right now.

Key Questions Answered

• •AI Capability Transitions: Four distinct leaps define the current AI era: GPT-3.5 in November 2022, GPT-4 in 2023, reasoning models peaking with o3 in early 2025, and workable agentic systems arriving December 2025. Recognizing which phase your organization operates in determines which tools and workflows are actually available to deploy today.
• •Agentic Work Restructuring: StrongDM's three-person software factory demonstrates a radical new operating model where human engineers spend $1,000 daily on AI tokens, with agents autonomously writing, testing, and shipping production code. Humans only review finished products. This signals that organizations should experiment with agent-based workflows now, before competitors establish precedents.
• •Rolling Disruption Pattern: As AI crosses capability thresholds, it triggers sudden market reactions, job announcements, and policy conflicts simultaneously — as seen in a single February week involving Citibank research, Block layoffs, and the Pentagon-Anthropic conflict. Organizations should build scenario planning processes that account for overnight perception shifts, not gradual transitions.
• •Recursive Self-Improvement Timeline: Anthropic, OpenAI, and Google DeepMind are all actively working to close the recursive self-improvement loop, where AI builds better AI. OpenAI's Codex was described as instrumental in creating itself. If this loop closes, the already-steep exponential capability curves accelerate further, compressing the window for organizations to adapt strategically.
• •Shape the Precedent Now: No established rules exist for AI use in workplaces, schools, or government. Every organization that figures out a responsible, effective AI workflow today sets a precedent others will follow. The actionable move is to run structured AI experiments internally now — not wait for regulation — because the window to influence norms is open but finite.