The 4Ds for individuals: how to actually get good at working with AI.
Most AI advice aimed at individuals is a list of prompt templates. Useful for an afternoon, useless for a career. Anthropic's 4D Framework is the opposite: a durable mental model you can apply to any tool, any model, for years. Here is how each competency plays out for one person doing real work.
Delegation — choose the work
Before you write a single instruction, decide whether AI should be involved at all, and in which mode. Anthropic describes three: automation (it does a defined task on your instruction), augmentation (you work turn by turn as collaborators), and agency (you set it up to run on its own within limits).
Summarising a video you just watched is automation. Asking AI to critique a memo you wrote is augmentation. Sending it off to research a topic and return a full report is agency. Naming the mode is half the skill.
Description — say it well
This is where prompting lives, but description is bigger than clever phrasing. It is giving context, constraints, examples and a clear definition of done. Most underwhelming output traces back to underwhelming description, not a weak model.
Discernment — judge it hard
The fluent individual treats every output as a draft from a fast, confident, occasionally wrong colleague. You check the reasoning, not just the vibe. You notice the plausible-sounding citation that doesn't exist, the edge case it skipped, the subtle drift from what you asked. Discernment is the competency that separates people who scale with AI from people who get quietly embarrassed by it.
Diligence — own it
You remain accountable. That means being transparent about AI's involvement where it matters, handling data responsibly, and never shipping something you can't stand behind. AI did the typing; you did the work.
Get good at one D per fortnight. In two months you will have moved further than a year of collecting prompts.
Want to know which of the four is your weakest right now? That is exactly what our assessment is built to surface.
Content informed by Anthropic's AI Fluency resources. Further reading: Anthropic — AI Fluency Framework ↗ · Anthropic — Claude for personal ↗