Most of the commentary on AI and work focuses on displacement: which jobs will be automated, which professions are safe. That's a real question, but it misses a more immediate and consequential shift happening right now — AI is fundamentally changing the economics of cognitive tasks.
Research has changed
The most dramatic shift I've observed in my own practice is in research. Tasks that previously required hours of document review — synthesizing literature, identifying comparable cases, understanding regulatory landscapes — can now be done in minutes. This isn't just a productivity gain. It changes what's possible.
When research is cheap, you can ask more questions. You can stress-test assumptions before committing to them. You can cover more ground before choosing a direction. The bottleneck has moved from gathering information to knowing which questions to ask.
That turns out to be a much harder skill to automate.
Expertise isn't going away, but it's being reconfigured
There's an argument that AI undermines expertise — that if a model can produce a competent market analysis in minutes, why pay for an analyst? I think this gets the causality backwards.
AI accelerates baseline competence, which raises the bar for what genuine expertise means. When everyone has access to good-enough analysis, the differentiator becomes judgment: the ability to identify the right framing, to challenge assumptions, to know what the data can and can't tell you.
Domain expertise — the kind that comes from years of pattern recognition in messy, real-world contexts — becomes more valuable, not less, because it's the lens through which AI-generated outputs get evaluated and applied.
The risk of confident mediocrity
The failure mode I'm most concerned about isn't AI replacing humans. It's humans outsourcing their thinking to AI and losing the capacity for critical evaluation.
AI systems produce confident output. They don't naturally surface their own limitations or flag when a question is poorly formed. Users who treat AI output as authoritative rather than as a starting point will consistently make worse decisions than those who engage with it critically.
This is the new literacy gap: not who can use AI tools, but who can think rigorously about the outputs they produce.