The Shift from Prompting to Demonstrating
AI interaction is shifting from explicit text prompts ("telling") to direct workflow demonstrations ("showing"). Driven by recent feature rollouts from major AI developers—such as OpenAI’s Record & Replay and Anthropic’s Record a Skill—systems are now designed to observe user actions, record screen workflows, and process spoken rationale to automate multi-step tasks autonomously.
Key Drivers of the Shift
- Tacit Knowledge & Context Gaps: Prompting fails to capture "tacit knowledge"—unspoken habits, edge-case rules, and personal preferences (e.g., manager approval thresholds on expense reports). Demonstrations capture these implicit decision points naturally.
- Reduction of "Botsitting": Users spend significant time correcting AI outputs and re-explaining context. Recording a workflow once eliminates repetitive back-and-forth prompt iterations.
- Workflow Mining & Scalability: Captured demonstrations enable organizations to build shared libraries of adaptable workflows, functioning similarly to open-source code repositories that refine over time.
Comparison: Prompt-Based vs. Demonstration-Based AI
| Feature | Prompt-Based Approach | Demonstration-Based Approach |
| Primary Input | Written text instructions | Recorded screen activity & spoken narration |
| Context Capture | Limited to explicitly stated text | Captures implicit workflows, tools, and edge cases |
| Efficiency | High manual effort; frequent re-prompting | One-time setup; runs autonomously in the background |
| Execution | Reactive (requires direct trigger) | Proactive / Scheduled (runs without manual reopening) |
The next phase of AI productivity relies on treating AI like a new employee: showing it how a job is done once through direct demonstration, allowing the system to retain organizational context and execute repetitive tasks without continuous supervision.