Lesson Brief
AI pioneers warn that the jobs most exposed to large language models are the ones that happen entirely behind a keyboard: drafting, summarising, classifying, looking up rules, and writing routine code. If your role is mostly text-in, text-out, you are operating on the same surface where current models are improving fastest, and the timeline being discussed inside the labs is measured in a couple of years rather than a generation.
The mechanism is data availability. Models learn from the enormous archive of human intellectual output that already lives on the internet, which is why office work was the first domino. Robotics lags because there is no equivalent dataset for moving a body through a messy, unfamiliar physical world. The distinction worth holding in your head is routine knowledge work, where outputs are predictable, versus contextual work, where the value sits in physical presence, judgement under ambiguity, or relationships built over time.
The tradeoff in auditing is honesty versus comfort. It is tempting to label your work as creative or strategic because that feels safer, but a credible audit forces you to count the hours you actually spend on routine tasks. If eighty percent of your week is composed of work a competent assistant model could draft a first pass on, your exposure is high regardless of your title, and a transition plan needs to start now rather than after the layoff.
Core Takeaways
- List every recurring task you do in a typical week and time-box each one in hours.
- Tag each task as routine knowledge work, contextual judgement, or physical presence.
- Routine keyboard tasks face the shortest displacement timeline and the steepest cost curve.
- Robotics lags software automation because no internet-scale dataset exists for physical work yet.
- Job titles disguise exposure; the audit must be at the task level, not the role level.
- A role above fifty percent routine knowledge work needs an active transition plan within twelve months.
Practice
Open a blank spreadsheet and list every task you performed in the last working week, one per row. In the second column, estimate hours spent. In the third column, tag the task as Routine (predictable input to predictable output behind a keyboard), Contextual (requires judgement, relationships, or physical presence), or Mixed. Sum the hours per category and calculate the percentage. If Routine is above forty percent, circle the three highest-hour routine tasks and write one sentence per task on what a current AI assistant would still get wrong. That short list becomes the starting point for your defence strategy in later lessons.