I often come across questions in chats like “Show me the specifics, how exactly is AI being used?” or “Which functions are already replacing agents?” and it seems to me the answer is obvious: almost everything. But I understand that “everything” is too broad to be considered an argument. So here are some fresh data from Anthropic.
They developed a metric called “observed exposure,” which shows the difference between what can theoretically be automated with AI and what is actually being used today. The main takeaway is that we are only beginning to explore these possibilities.
What’s interesting: for programmers, about 75% of tasks are already covered. In the “computer & math” category, real application reaches roughly one-third — 33%, while the potential there is estimated at 94%. That means we are currently using only a third of what’s possible, even in the most advanced field.
Regarding office and administrative tasks — the theoretical coverage reaches about 90%. In finance, law, accounting, or HR, according to the study, the most susceptible to automation are people with high education levels and good salaries. Often, they are the ones asking: “And what functions?” — and the details are on the chart.
For example, almost all my analytics, financial modeling, investment research, or preparation for calls and deals are now handled by the tool Claude. It works faster, cheaper, and provides better quality than I could do manually.
Since the launch of ChatGPT, demand for young specialists in these areas has decreased by about 14%. There has been no increase in unemployment — companies simply stopped hiring new people. It’s not a crisis: rather a quiet replacement of old processes with automation. Moreover, growth forecasts for such professions until 2034 are less optimistic — the market considers automation a key trend.
The main issue is not whether agents can perform the work — data confirms they can. The much more important question is the speed of change for the people themselves. Take 3-5 of your key work functions and try running them through an agent this week. The result will surprise you more than any report. And if you want to understand how to do this yourself or within your company — just look for information on these topics.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby