+ Microsoft's CEO Called For A Price War On AI
In a Wall Street Journal interview published June 22, Microsoft chief executive Satya Nadella argued that the next phase of AI should be about making it cheaper, more trustworthy, and easier to swap, rather than more powerful. He criticized rival labs for telling the public that white-collar work is finished while asking for unlimited money to build data centers, and said the industry has to earn what he called the public's permission.
The price war is already showing up in Microsoft's own bills. Copilot Cowork, the assistant that does multi-step work on its own that we covered last week, now charges by usage rather than a flat monthly fee. It bills in small credits that cost about a penny each, so a company pays for what it actually uses, on top of the paid Copilot license each person still needs. Microsoft is also weighing whether to offer DeepSeek, an ultra-cheap Chinese model, inside Copilot, and it has built its own lower-cost models plus an agent that hands each task to whichever model is cheapest. Microsoft makes most of its AI money running other companies' models for them, so when models get cheap, Microsoft loses little and the labs that sell the models lose a lot.
What it all means: When the company that rents you the shelf space says the products on it should be cheap and interchangeable, notice who that helps. Cheaper, swappable models and pay-for-what-you-use pricing are good for the people buying them. They are also exactly what strips the value from the companies Microsoft competes with.
+ Anthropic Put Claude Inside Slack As A Full-Time Teammate
On June 23, Anthropic launched Claude Tag, a version of its AI that lives directly inside Slack channels and behaves like a permanent member of the team. Instead of opening a separate app, you tag Claude in a channel to hand it work, ask for a summary, or let it jump into a conversation on its own when it spots something it can help with. It is in beta for Anthropic's business and enterprise customers.
What makes this different from the chat bots companies already have is memory and reach. Each Claude has its own identity tied to specific channels, remembers past conversations, and can pull facts from other channels it has been approved to see, so a legal Claude cannot read engineering chats and the other way around. Hand it a multi-step job and it splits the work into stages, does them with the tools it has access to, and posts its progress in the thread for everyone to follow. Administrators decide what each Claude can see and when it is allowed to speak.
What it all means: The AI is moving out of its own window and into the place your team already talks all day. That makes it far easier to use and far harder to ignore, and it hands every manager a new problem: deciding what the assistant may read, and how anyone checks the work it did on its own.
+ Santander Said AI Will Save It More Than 430 Million Pounds
On June 22, the global bank Banco Santander said it expects to cut costs by more than five hundred million euros, about four hundred and thirty million pounds, by rolling AI out across its operations. The bank is giving all 185,000 of its staff worldwide access to AI tools and is targeting more than a billion euros in combined savings and new revenue between 2026 and 2028.
Some of the detail is concrete. In the UK, Santander is pointing AI at card-related phone calls, aiming to handle 240,000 of them, about 40 percent of the yearly total, through self-service. The bank has not said whether the savings will cost jobs, and has not announced any cuts tied to the rollout.
What it all means: Santander is one of the first big banks to put a hard number on what AI saves, and that number becomes a benchmark every competitor and board will measure against. Once a savings figure is public, the pressure to match it spreads fast. The line the bank did not draw, between cost saved and jobs kept, is the one its own staff will be watching.
