People treat the AI wave as one long event. It was not. It was a sequence of distinct literacies, each obsoleting the last as the default and stacking on top of it as a floor. If you learned the wrong year’s skill, you were fluent in a language no one spoke anymore. Here is the timeline as a curriculum, not a hype cycle.

One skill per year

• 2022: Reading the labs. Model cards, constitutional AI, the academic-feedback framing of Claude versus ChatGPT. The literacy was evaluative: knowing how a model was shaped, not just what it said.
• 2023: Development and deployment. Building and shipping went mainstream. The literacy was operational: prompts, context, fine-tuning, and getting a model into a product without breaking it.
• 2024: Choosing your surface. Public scale versus enterprise deployment. NotebookLM versus Copilot versus Glean. Cloud versus local. The literacy was architectural: matching the tool to the data, the trust boundary, and the bill.
• 2025: Preparing for the job market. New grads walked into a field that already assumed all three prior literacies. The literacy was strategic: positioning yourself where the work is going, not where it was.

The 2025 shift lands hardest in APAC. India, Pakistan, Nepal, Sri Lanka, Malaysia, and Indonesia are producing enormous cohorts of new graduates into a market where the entry-level rung has changed shape. The routine work that used to be the first job is the work most exposed to automation. That does not remove the opportunity. It moves it up a level, and it rewards whoever can direct the tools rather than compete with them.

If you are catching up now, do not start at 2022 and crawl forward. Start at the current floor and learn downward only when you hit something you do not understand. The point of a field guide is not to relive the trip. It is to skip the wrong turns.