If your company runs on Microsoft 365, the artificial intelligence answering a prompt inside Excel or Outlook is quietly becoming a different product than it was a few months ago. Bloomberg reported on July 7, 2026 that Microsoft has started routing some of those prompts to its own in-house models instead of OpenAI's or Anthropic's, and reported on July 15, 2026 that its sales organization is now being coached on how to explain that choice to buyers.
For a leader who feels a step behind on AI, this is a useful moment. You do not need to pick a side in a fight between three of the largest technology companies in the world. What makes it useful is that it shows you, in plain view, how vendor claims get made, and why knowing which model is doing the work is now part of running an AI-capable company.
What actually happened
Two pieces of reporting sit underneath this story.
The first, from Bloomberg on July 7, 2026, is that Microsoft has begun replacing OpenAI and Anthropic models with its own inside some Microsoft 365 apps. Excel and Outlook are the two apps confirmed so far. The volume is real but still small: reportedly tens of thousands of prompts a week are now handled by Microsoft's in-house models, a modest share of total usage.
The second, from Bloomberg on July 15, is that Microsoft held an internal sales strategy meeting the week of July 15 to prepare its people for the new fiscal year, FY27. According to Bloomberg's account, executives outlined a plan for salespeople to compare rival AI products from OpenAI and Anthropic unfavorably to Microsoft's own on cost, security, and product completeness.
A note on how to read all of this: the internal-meeting details come from Bloomberg's reporting, and Bloomberg's original article is paywalled. The quotes and meeting specifics are consistent across multiple outlets covering the same story, but they are reported claims, not something anyone outside that room has confirmed on the record. Microsoft and Anthropic had not issued an on-record response at publication. Treat the meeting details as reported, and hold them a little more loosely than the underlying product facts.
The pitch: full system versus parts
The clearest window into Microsoft's thinking is a line attributed to Jay Parikh, the Executive Vice President who leads Microsoft's CoreAI platform and tools organization and reports directly to Satya Nadella. According to Bloomberg, Parikh told the sales teams:
Everyone else is selling parts. We're selling the full end-to-end system. That's the story that we all need to get out there and tell in FY27.
That one sentence is the whole strategy. Microsoft owns the models, the apps, the security stack, and the cloud underneath. Its argument to a buyer is that owning the entire system beats assembling best-of-breed pieces from different vendors.
Jacob Andreou, the Executive Vice President for Copilot, reportedly went more specific. Bloomberg's account has him describing Anthropic's Claude model inside Microsoft's Office apps as slower, less accurate, and lacking the proper security integrations.
The reason underneath: cost
The product argument sits on top of a money argument. Back in the July 7 reporting, Mustafa Suleyman, the Executive Vice President and CEO of Microsoft AI, was reported to have described the motive as reducing, and ultimately eliminating, what Microsoft pays Anthropic, a detail from paywalled Bloomberg reporting we could not independently verify. Treat it as reported rather than confirmed, but the direction is consistent with everything else here.
Microsoft has been building toward this for a while. At its Build 2026 developer conference on June 2, it launched seven of its own MAI models spanning reasoning, coding, image, transcription, and voice. Its flagship, MAI-Thinking-1, runs about 35 billion active parameters with a 256K-token context window. Microsoft says independent raters preferred it to Anthropic's Sonnet 4.6 in a blind test, and that it matches Anthropic's Opus 4.6 on a coding benchmark called SWE-Bench Pro. Those are Microsoft's own published claims about its own models, which is worth keeping in mind when you read them.
In-house MAI models Microsoft launched at its Build 2026 conference, spanning reasoning, coding, image, transcription, and voice.
Source: Microsoft AI, 2026Why this affects you, even if you never see the switch
Most business leaders will never notice which model answered a given prompt. That is exactly why this is worth a minute of your attention.
The tool you rely on can change underneath you
When a vendor swaps the engine inside a product you already pay for, your experience can improve, hold steady, or drift, and you have no visible switch to check. The practical takeaway is awareness. Knowing that the AI inside Outlook is not one fixed thing, and that it can be a different model this quarter than last, is now a basic part of being AI-capable. It belongs in the same mental file as knowing which cloud your data sits in.
Vendor claims are sales claims, and discernment is the defense
This story is a clean example of the difference between a fact and a pitch. That Microsoft owns the models, the apps, and the cloud underneath is a fact. That its rivals are slower, less accurate, and less secure inside Office is a pitch, and one Microsoft is reportedly coaching its people to deliver. Some of it may hold up under testing. Some of it is positioning. You cannot tell which from the sentence alone.
The skill that protects a buyer here is the same one that protects anyone using AI at all: trust, then verify.
In The 7 Levels of AI Proficiency, this is the Level 3 muscle, the Critical Thinker. A Level 3 leader does not take a vendor's comparison at face value and does not reject it out of reflex either. They ask for the test. For your own company, that means running your real workflow against the options that matter to you, on your own data, and judging the output yourself rather than trusting the slide.
Own the system or buy the parts is your question too
Microsoft's full-system pitch is a version of a decision every owner faces with their own AI stack. Do you commit to one vendor's whole ecosystem for the convenience of one throat to choke, or do you assemble the best individual tools and accept the work of stitching them together? There is no universal right answer. There is only the answer that fits your team's capability, your data, and your tolerance for being locked in. Microsoft is betting most buyers will pick the single system. Whether that is right for you is a judgment call, not a default.
Cost is now a first-class reason to choose a model
The quiet lesson in the swap is that Microsoft is changing models partly to spend less, not only to perform better. That same logic scales all the way down to a small business. The most capable model is not always the right one for a given task. The cheapest model that does the job well enough often wins, and matching the model to the work is a competence worth building, not a corner worth cutting. A leader who understands this treats model choice as a line item they can manage, the way they already manage every other tool budget.
Your next step
Pick one AI task your team does often, inside whatever tool you already pay for. Run it, read the output with a critical eye, and ask whether it actually did the job. That single habit, repeated, is how a company moves from taking vendor claims on faith to judging them on evidence. It costs you a few minutes and it is the most practical way to build the discernment this whole story rewards.
Related reading: Level 3: The Lieutenant (Critical Thinker).
Sources
- Microsoft is reportedly training salespeople to talk down OpenAI and Anthropic (TechCrunch)
- Microsoft Gives Salespeople Tips to Knock Down Anthropic, OpenAI (Bloomberg Law excerpt)
- Microsoft Replaces OpenAI, Anthropic With Own AI in Some Apps (Bloomberg)
- Building a Hillclimbing Machine: Launching Seven New MAI Models (Microsoft AI)
- Announcing Copilot Leadership Update (Microsoft)
- Microsoft Build 2026 (Microsoft)
- Microsoft unveils new AI models to lessen reliance on OpenAI, lower costs (CNBC)
Frequently Asked Questions
Does this mean Copilot is worse now?
There is no evidence of that. Microsoft is routing some prompts to its own models and claims its flagship performs at or above rival models on the benchmarks it published. The honest answer is that performance depends on your specific tasks, which is why testing your own workflow beats trusting any single claim.
Which apps are affected?
The reporting confirms Excel and Outlook as the apps where Microsoft has started using its own models. Other Microsoft 365 apps have not been confirmed in this round of reporting, so it is best not to assume beyond those two.
Should I switch vendors because of this?
Nothing here calls for a reaction. It calls for awareness. The useful step is to know which tools your team depends on, understand that the models inside them can change, and build a simple habit of verifying vendor comparisons against your own results before you act on them.
Find your AI Proficiency level
The free 7 Levels assessment places you across seven stages of AI capability. Under ten minutes. Research-backed scoring.