What happened
On 13 August Microsoft announced that it is merging its two Copilot apps — the consumer one and the Microsoft 365 one — into a single app, and that several features are being switched off along the way. The first retirements begin on 18 August. Five days sit between the announcement and the deadline.
The list, starting with the sharpest item:
Podcasts. The audio-generation feature is being retired in the Copilot app for consumers from 18 August, on free and paid accounts alike. After that date, previously created podcasts stop being available, and any links somebody shared to them stop working.
Group Chat. Shared conversations with Copilot are going away; account updates begin on 18 August, with the exact moment depending on the account. Your own messages, prompts and artifacts stay. Messages and artifacts from the other participants will not be accessible in the migrated chat — Microsoft explicitly advises downloading what other people contributed beforehand.
Copilot Labs. Content generated inside the experiments does not survive the account update. The timing varies by account, so there is no single date to put in the calendar here.
Mico. The animated character in voice mode is being removed in waves from August. Voice conversations remain in chat history; the character settings do not.
Consumer Deep Research. Here we are relying on trade-press reporting rather than the vendor documentation we could confirm for the other items: the feature is said to disappear on the same date, with Microsoft 365 Premium subscribers getting a tool called Researcher in its place.
One thing in this story was reported wrongly, and it happens to be the one that governs what you have to do before 18 August. A good deal of the coverage said podcasts cannot be exported. Microsoft's own documentation says otherwise: on Copilot for web, an individual podcast can be downloaded via the Download option in the library menu. The library goes away and other people's links go away — the ability to download does not, as long as the deadline has not passed.
It is worth narrowing the blast radius straight away, because the headlines do not: these retirements affect the Copilot app for consumers, not an organization's Microsoft 365 Copilot deployment. Except that in plenty of mid-sized companies that free app is exactly where people actually work with a model, because nobody bought them anything else.
Our take
This is not a vendor slip-up or a broken process. It is the process — which is what makes it more interesting than the feature list itself.
Notice what is missing from this story. There is no price increase, no change of contract terms, not even a drop in quality. What there is instead is something risk registers rarely have a row for: a feature somebody built their way of working on stops existing on a date set by somebody else. A price can be negotiated. A retirement date cannot.
In June we wrote that the cost of a tool stopped being a fixed line item when GitHub Copilot moved to usage-based billing. This is the same layer of risk, moved one step further: back then the price of what you have changed, now what changes is whether you have it at all.
The practical conclusion is narrower than it might seem, and it is not "don't use features that come inside a suite". It is this: export is a feature you need to have before you need it. Five days is enough time to pull a library's contents out — provided somebody in the company knows that library exists and who uses it. If nobody knows, five days is irrelevant, because the clock is running past you rather than for you.
Hence a distinction worth writing down before the next announcement of this kind arrives: what is convenience at your company, and what is load-bearing. Convenience can disappear on a Friday and nobody will note it. Load-bearing is whatever goes into client material, into a decision, or into documentation — and that has to have its own storage and its own way out, however well it works inside the suite. This is how we lay that layer out in our services: first a list of the places where a process rests on somebody else's feature, then, for each one, an answer to where the data sits and what you get it out with.
And one thing we flag explicitly as our expectation rather than an established fact: app consolidation rarely ends after one round. Microsoft has announced that it is merging two Copilots, which usually means further features on both sides will overlap and some of them will go the same way. We do not know which, or when. We do know that the answer to "what do we pull out" is better prepared before the day of the announcement than on it.
Why this matters
Private Equity
For a fund this is not a deal-level risk item but a cheap test of operational maturity inside a company. One question at the next review exposes the state of things: if a vendor switches off a feature the team uses tomorrow, who finds out, and how? A company where somebody reads vendor product notices and keeps a list of dependencies is in a different place from one where users learn about a retirement on the day it happens. After a few acquisitions the second situation is the norm, because the tools arrived with the teams and nobody ever wrote them down.
Enterprise
A large organization has vendor management, and usually has it written around contracts rather than around features. Retiring a single feature typically breaches nothing in the agreement — and can still stop somebody's work. The detail worth noting comes from group chats: after migration your part stays and everybody else's disappears. In practice that means shared material has a different owner than it appears to, and that "it's in Copilot" is not an answer to the question of where something is.
The second thread is tool use outside the official deployment, and not in a pejorative sense. The consumer app often runs alongside the corporate Copilot — not out of bad faith, but because it is at hand. If that is the case at your company, this announcement concerns your people even though it formally does not concern your contract.
SMB / mid-market
The impact lands most directly here, because in a mid-sized company the free app often is not an add-on but the entire AI deployment. None of this means you have to buy something. It means settling two things this week: whether anyone at your company made podcasts or worked in group chats, and whether any of it went out to a client. If it did, that is material which after 18 August will exist only where you saved it yourself.
One move this week
Before 18 August, put one question to the team and give them one day for it: which Copilot features do you actually use, and does any of them hold something that does not exist in a second place? Do not accept "probably nobody" without checking — the podcast library and group chats are exactly the places where things get left without a second thought, because they look like part of the system. Whatever you find, download it now: the download option works up to the deadline, not after it. The whole job is an hour, and the alternative is reconstructing something that no longer exists. Describe your case: mailto:[email protected]?subject=Rozmowa%20z%20Aurora%20AI.