The AI Job Shift

White-collar disruption, plus practical tools for AI-powered development and design

An article I read recently got me thinking about the future of work and what we need to do to prepare for it. Anthropic CEO Dario Amodei is sounding the alarm that AI could eliminate half of all entry-level white-collar jobs within 1-5 years, potentially spiking unemployment to 20%. Amodei warns that companies are already racing to develop AI agents that can replace human workers in coding, finance, law and consulting, not just assist them. The shift from augmentation to automation could happen suddenly, with businesses mass-replacing workers once AI reaches human-level performance.

So what can we do about this? One thing is to start to understand this technology better. Luckily, Anthropic has just released this excellent short course on their AI Fluency Framework. This 12-module course teaches effective AI collaboration through a "4D Framework": Delegation (knowing what to assign to AI), Description (clear communication with AI systems), Discernment (evaluating AI outputs critically) and Diligence (maintaining ethical and safe practices). The course covers the fundamentals of generative AI, advanced prompting techniques, and three key collaboration modes: automation, augmentation, and agency.

If you can’t beat ‘em, join ‘em - working with Agents

I’m a project-based learner, so I have been exploring the use of agentic tools to determine the most effective way to work with them.

The first one I’ve used is Marnus, a general-purpose AI agent. This project tackled the challenge of explaining large language models to 11-year-olds through an interactive browser game called "How AI Thinks." With a detailed brief outlining the features, Manus generated complete working code in roughly 30 minutes. While requiring several rounds of refinements, the foundation was solid and functional. The project demonstrates AI's practical potential for rapidly developing educational content, but my takeaway from using Marnus is that it was too much automation and not enough human input. I think we could have ended up with a better product if we had more opportunities for collaboration.

Here’s the game.

The second project involved working with Cursor’s new agentic mode, along with a structured workflow suggested by Ryan Carson. This project built a complete AI literacy assessment tool for teachers based on the UNESCO AI Competency Framework. Instead of just "vibe coding," the approach started with a comprehensive Product Requirements Document (PRD) that Cursor AI converted into 43 specific tasks across six phases: project setup, survey interface, scoring system, results dashboard, responsive design and deployment. Using the Cursor IDE systematically, the entire production-ready web application was built in a matter of hours. The key insight: high-quality context (through the detailed PRD) and ongoing human collaboration dramatically improved the AI output quality.

Here’s the app.

Cool Tool

Stitch by Google is an AI design tool powered by Gemini that lets non-designers create web or mobile UIs through prompts or image uploads, then export to Figma or HTML code, making it valuable for anyone who has more ideas than design skills.

See you next time. Niall