- Claude is used much more by developers than by customer services, sales, finance, or e-commerce.
- It now works more autonomously than ever, including complex tasks.
- Claude clarifies details more than humans interrupt.
Anthropic has revealed that around half of all calls to public API tools come from software engineering, while other sectors such as customer service, sales, finance and e-commerce only account for a few percentage points each.
Additionally, Claude Code was found to run autonomously for longer than ever, now running at over 45 minutes, compared to less than 25 minutes just three months ago.
OpenAI says the increase in autonomy is not just related to its latest high-performance models, suggesting a broader shift toward agent AI rather than testing of its newer models.
Agent AI is here and it’s more common in coding
In a new blog post, OpenAI described how users are starting to build trust with AI as tasks become more ambitious; however, there is still “over-deployment” whereby models are even more capable than users currently allow. External evaluations also show that the models can perform much longer tasks under ideal conditions.
Claude is also helping on the trust front, stopping to ask clarifying questions more than the humans interrupt.
In terms of coding, it seems that the more complex the task, the less human supervision there is. Complex tasks, such as discovering exploits and writing compilers, receive about 67% human oversight, compared to 87% oversight for simple API tasks.
Claude also asks about twice as many questions about very complex tasks as he does about simple tasks.
Looking ahead, Anthropic covered the importance of training models to recognize uncertainty and proactively ask clarifying questions. The company also advises against requiring manual approval for every action, saying it adds friction without increasing the necessary security.
Follow TechRadar on Google News and add us as a preferred source to receive news, reviews and opinions from our experts in your feeds. Be sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp also.
