The generative frontier is where the pressure is highest now
Writing, images, code and design moved first and fastest.
Not all tasks are equally exposed, and the order is not random. The work that turned out to be easiest for current AI is the production of plausible content: text, images, layout, boilerplate code. If your value is mostly in generating a first draft of any of these, you are standing closest to the fire. That does not mean run; it means know where you are.
What the evidence shows
One widely cited estimate: current generative AI and other technologies combined could automate activities absorbing 60 to 70 percent of employees' time, with roughly three quarters of generative AI's value landing in customer operations, marketing and sales, software engineering and R&D.
Writing is the single most exposed knowledge task; real usage data shows writing and coding as the two biggest categories of AI use.
Goldman Sachs put the exposure of generative AI at the equivalent of 300 million full-time jobs, weighted heavily toward white-collar content work.
The frontier moves. Tasks that look safe because they are physical or relational today can be reached later as robotics and voice improve. 'Safe for now' is a real category, but it has a date on it.
If your craft is content production, shift your value up the stack: toward judgment, direction, accuracy and the things a draft cannot give you.