AI moves into customer service, legal and accounting, but the bill is coming

AI is no longer just a chatbot you ask for advice. In this week's stories, it becomes part of the workflows themselves: it answers the phone, reads contracts and pulls financial figures.

Starlink is using xAI's voice model Grok Voice for about 20,000 customer calls a day, and according to the figures provided, roughly 70 percent end without a human employee. The figures have not been independently verified, however. Microsoft has launched Copilot for Legal, which is meant to help lawyers review and write contracts, among other things. And Xero users can now pull live financial figures directly into Google's Gemini.

More places to use AI, and more places to go wrong

At the same time, AI is being connected to money. The broker Axi has linked its trading platform to Claude, so the assistant can show accounts and prices and place orders when the customer approves. It is a practical detail that matters: a human still has to say yes.

Threats come with it. World, which Sam Altman co-founded, has launched a package for banks designed to detect deepfakes in video calls, meaning fake, AI-generated faces. No bank has yet been named as using it.

The bill and the capacity

When AI becomes an everyday tool, so does its price. According to internal messages cited by 404 Media, from October 1 Accenture will fund only $29 per employee per month for Claude, where some previously used up to $1,000. Pricing models are also spreading: Meta offers a free basic version of its agent Muse, while OpenAI's Dots requires a subscription starting at $100 a month. An agent is an AI that keeps working on a task in the background.

On the company side, the numbers point upward. According to the Financial Times, OpenAI told investors that its annualized revenue (the figure extrapolated to a full year) was close to $50 billion at the end of September. According to Chinese media, DeepSeek has doubled its funding round to around 100 billion yuan. But DeepSeek's chat, app and API were down for almost five hours on October 15. The cause has not been disclosed, but media and experts point to pressure on computing power.

Politics is also getting involved. China has put a draft of export rules out for consultation that could hit the largest and most advanced models, but not ordinary use through apps and websites.

What can you use this for?

The pattern is clear: AI can do more, but every task comes with a price, a risk and a dependence on a few providers. It is a good time to look at which of your tasks actually save time, and what they cost per employee, as Accenture's cap suggests.

Also watch whether banks begin to name their use of deepfake protection, whether Starlink's figures are verified, and what ends up in China's final export rules. Where AI sits close to finances and contracts, it is an advantage to have a human approve the final step.