Microsoft just put $2.5 billion into helping you use AI (and other AI business news)
getty
Getting a business to do something useful with AI is hard. There is more AI news than anyone can read, and more a company could do with it than any team has time to test. The difficulty is knowing which of it will change anything for your clients and your staff. Get that judgment right and the tools pay for themselves. Get it wrong and you waste months on the wrong thing.
July brought some AI news that business owners should pay attention to. Two of the biggest companies in the world spent billions on the same bet inside 48 hours. OpenAI released an agent that disappears for hours and comes back with the whole job done. For the first time, the AI industry got graded on safety, and the company that ranked first still fell short. A restaurant can now take your order from inside a chat window. And Anthropic found a secret part of its own model, a place where it works things out before it says a word.
Here’s the AI news and what it means for your business.
The AI news from July that changes what you build and sell
Microsoft spent $2.5 billion helping companies use the AI they already have
On 2 July, Microsoft launched Frontier Co., a $2.5 billion unit staffed with around 6,000 engineers, consultants and specialists who go into client companies and work next to their teams to get AI running inside the daily work. Two days earlier, Amazon put $1 billion into a matching effort inside AWS, sending small groups of engineers to work inside a customer’s business for up to 45 days at a time.
Both companies looked at one bottleneck and drew one conclusion. Too many business owners buy the subscription and never change how their business runs. But we know companies that use AI in their service are worth 130% more. Your service is where the growth is. Learn AI. Raise your prices.
OpenAI’s new agent works a task for hours and returns the finished file
On 9 July, OpenAI released ChatGPT Work, an agent built on its new GPT-5.6 model that gathers information across your connected apps and files, breaks a task into steps, and keeps at it on its own for hours to produce a finished document, deck or report. It checks in only when it needs a decision from you. Scheduled tasks keep projects moving without anyone opening the app.
An agent that spends hours on a full deliverable used to need an enterprise contract. Now the free tier gets it too. Pick one report you build every week, give the agent the same brief you would give a junior hire, and let it run while you take the call only you can take.
Restaurants started taking orders inside ChatGPT and Claude
On 1 July, Square rolled out new integrations that let US food and drink sellers get found and take orders directly inside a conversation with ChatGPT or Claude, with no technical setup. Orders route straight into the seller’s existing point-of-sale and kitchen screen. Square adds no marketplace commission on top of its usual online ordering fee, which runs around 2.9% to 3.3% plus 30 cents per transaction. Alexa is next.
This is the first clear case of AI chat working as a sales channel for a small business. Until this month it drafted your emails and summarized your documents. Now it takes payment. If you sell anything direct to consumers, watch this one closely. The question to answer is whether your product can be bought inside the same conversation where your customer is already asking the AI what to order.
The safest AI company in the industry scored a C+
On 7 July, the Future of Life Institute published its summer 2026 AI Safety Index, graded by seven outside reviewers across 37 measures. Anthropic ranked first of nine major AI companies and only came away with a C+. OpenAI and Google DeepMind each got a C. xAI, DeepSeek and Mistral failed outright. The reviewers found that several firms have softened earlier promises to pause development near set danger thresholds.
The people asking the world to run sensitive work through their tools admit their own safeguards are behind. Your value is being the person who checks the output, catches the miss, and takes responsibility for the result. Use the tools, but don’t miss out the human judgement. Clients pay you to own the outcome.
Anthropic found where its model works things out in private
On 6 July, Anthropic published new research identifying a small part of Claude, which it calls J-Space, that keeps the ideas the model works through before any of them appear in the output. In one test, the team told Claude to think about the Golden Gate Bridge while copying an unrelated sentence, and found “bridge” and “California” active in that part the whole time, even though the visible answer never mentioned either. In a version trained to sabotage code, the words “fake” and “fraud” turned up there before the output looked normal.
What a model says and what it works through can differ. Now there is a way to see the gap. When you let AI suggest something that is important to your client, do not stop at the answer on the screen. Read the reasoning behind it, and confirm the two line up. Don’t trust the output without looking underneath.
Why Microsoft and Amazon are spending billions to help companies use AI
The companies building AI have decided that using it well is the hard part, and they are putting serious money behind that call. An agent now handles a full job on its own. Inside a chat window, a customer places the order and pays. Underneath a confident answer, a model can hide a problem you would miss. Your opening is the judgment on top of all of it, knowing what to hand a tool and what to keep in your own hands.
Get my free AI playbook for ambitious founders looking to scale.
