Product
Anthropic, creators of Claude, have released a report that measures how autonomous AI agents are being implemented. One of their key findings is that agentic use of their models is concentrated on just one domain: software engineering. It accounts for almost 50% of tool case use of their API. Travel, legal, and healthcare each represent 1% or less of AI tool calls, while e-commerce, education, and document creation each represent less than 2% of AI tool use.
The data suggests that we are in the very early stages of agent adoption. AI agents are being severely underutilized in many domains, and early adopters may find considerable value to gain. Identify tasks that are sufficiently complex, somewhat expensive, and where the cost of errors is low. These are tasks where AI agents can deliver tremendous value. Software engineering is one such example.