Descriptions:
Dylan Davis, who runs an AI consultancy, addresses one of the most common mistakes he observes clients making: letting AI build dashboards and full applications when a simpler solution already living inside the AI workspace would do the job better. The video introduces a three-tier decision framework for placing AI workflows: inside the AI workspace itself via projects or skills (estimated to cover 85-90% of use cases), as a rendered artifact or view the AI generates for you (5-10%), or as a standalone hosted application (roughly 1% of cases).
Davis unpacks the distinction between projects and skills within Claude and ChatGPT: projects are focused workspaces that prime the AI for a recurring activity, while skills are portable prompt templates callable across any conversation. He notes that Claude supports both in the browser while ChatGPT currently supports only projects, and warns against accumulating too many overlapping skills — recommending a cap of around 15 for browser-based Claude users to avoid the model selecting the wrong one due to similar titles or descriptions.
The third tier, live artifacts in Claude’s desktop co-worker client, is also demonstrated. Unlike standard browser artifacts, live artifacts support live data connectors to external systems like CRMs, task trackers, calendars, and email — enabling dashboards that update automatically. The central reframe throughout is shifting from “can AI build this?” to “where should this live?”, a practical mental model for reducing unnecessary maintenance burden in AI-assisted professional workflows.
📺 Source: Dylan Davis · Published May 13, 2026
🏷️ Format: Deep Dive
