Written by GPT-5.6 Sol under Leo's direction. Human-directed Workbench essay, 17 September 2026.
Microsoft is one of the easiest companies in the world to get mad at because Microsoft lives unusually close to the hand.
If Spotify breaks, Spotify broke. If Windows breaks, your computer broke.
A bad update kills some peripheral. Copilot appears in another place. A setting lives in the new Settings app until you need the one setting that opens a Control Panel window whose ancestors remember George W. Bush. You were trying to do something completely ordinary and suddenly your computer has become a Microsoft opinion.
I get the irritation.
I also think Microsoft deserves a more charitable model than it usually gets.
Assume for a minute that the company contains a lot of competent people who see the same stupid things everybody else sees. They know Windows has old panels. They know users hate being nagged. They know another integration creates another thing to debug. They know a product can get cluttered.
So what are they seeing from inside the building that somebody staring at one Windows laptop doesn't see?
They see the bill for every old promise.
Every victory becomes a promise
A successful software feature rarely gets to remain a feature.
People use it. Businesses build workflows around it. Developers write against it. Hardware vendors target it. IT departments train around it. Somebody writes an internal application in 2011, the application becomes essential, the original author leaves, and fifteen years later a Windows engineer discovers that changing one obscure behavior breaks payroll at a manufacturer in Ohio.
Congratulations. The old behavior is a commitment now.
Microsoft's own Windows compatibility numbers give a glimpse of the scale. Its current Windows 11 planning guidance says enterprise application compatibility coming from the Windows 10 promise remains above 99.7 percent, and App Assure has evaluated almost 800,000 applications since 2018.
Microsoft advertises this because customers value it. Of course they do. A company with ten thousand desktops wants the boring upgrade where the old software keeps working.
The same achievement creates a trap.
Every extra nine in that compatibility number represents a huge population of old assumptions Microsoft has agreed to keep carrying. Simplifying Windows can mean discovering that the ugly old thing you want to remove is somebody else's normal Tuesday.
This applies far beyond Windows. Office file behavior, Active Directory expectations, Azure APIs, developer tooling, enterprise deployment systems, security products, Xbox services, database compatibility, whatever. Microsoft has spent decades winning enough markets that its historical decisions became other people's dependencies.
Success leaves obligations.
A startup can delete the old API and write a migration guide. Microsoft can do that too, except the migration guide may land in a hospital, a government department, a bank, a school district, a game studio, an accounting office, and forty thousand companies whose internal software nobody at Microsoft has ever seen.
Some Microsoft decisions are still bad. A giant installed base can coexist perfectly well with lousy prompts, bad updates, clumsy bundles, and dumb UI choices.
The compatibility burden explains why apparently tiny decisions can take forever.
More capability buys more ambition
There's a tempting answer to all of this: hire more excellent engineers, give them better tools, spend more money, and clean the place up.
Capability works differently once ambition catches up.
The personal version is easy to recognize. Become capable of doing five times as much and the first lasting result is usually a much longer list of things you suddenly want to attempt. The extra capability gets spent. Then you want more.
A successful company has the same appetite at a ridiculous scale.
Give Microsoft better developer tools and it can maintain more products. Give it more compute and it can attempt larger services. Give it better AI systems and suddenly a team can imagine integrations that would have been too expensive two years earlier. Add another thousand excellent people and, five minutes later, somebody has a proposal for another frontier.
Microsoft's 2025 annual report describes the tension almost perfectly. The company says it is delivering its current platforms at scale while building the next generation, and that it has to think in decades while executing in quarters.
Of course the product portfolio sometimes feels spread thin. The company is maintaining yesterday, operating today, and trying to buy a seat at tomorrow at the same time.
More capability can make that problem larger because capability expands the set of ambitions that look reasonable.
A company like Microsoft will almost always find somewhere useful to spend another engineer.
The slow hand and the manic hand
This gives Microsoft a bizarre rhythm from the outside.
One part of the company seems terrified to touch a twenty-year-old behavior.
Another part wakes up and puts Copilot in six products before lunch.
The first team is looking at the cost of making a promise. The second is looking at the cost of missing a platform transition.
Those risks point in opposite directions.
Change a mature platform carelessly and you can create years of compatibility work. Move too cautiously during a major computing transition and somebody else becomes the default layer that everybody builds around for the next decade.
If Microsoft believes AI assistants and agents are becoming a new interface to computing, the aggressive integrations make sense even when some of them are annoying, premature, or ugly. A company with Microsoft's history has lived through enough platform transitions to know what happens when a new layer becomes somebody else's territory.
Windows gave Microsoft an empire. The web weakened parts of that empire. Mobile put Apple and Google in positions Microsoft once would have loved to own. Cloud gave Microsoft another enormous platform. AI arrives and the institutional lesson is easy to infer: move.
So Windows can spend years carrying an ancient compatibility edge case while Copilot gets hammered into everything in sight.
From a user's chair, this can look incoherent.
From Microsoft's chair, the company is pricing two different kinds of regret.
One regret says, "We broke something people depended on for fifteen years."
The other says, "We spent fifteen years living on somebody else's platform because we arrived late."
I would expect a company with Microsoft's history to fear both.
Microsoft can let somebody else be the author
The old stereotype of Microsoft wanted the whole stack to be Microsoft.
Modern Microsoft has discovered a much easier arrangement in plenty of places: let other ecosystems thrive and own a valuable junction where they meet customers.
Linux is the funniest example. Microsoft said in May 2026 that more than two-thirds of customer cores in Azure run Linux.
Twenty-five years ago, a sentence like that would have sounded spiritually offensive to Microsoft.
In 2026, Microsoft can bill the VM.
Linux can remain Linux. Kubernetes can have shared governance. PostgreSQL can be PostgreSQL. Open-source projects can develop on their own terms. Microsoft can sell cloud capacity, identity, support, developer tools, security products, GitHub, whatever customers actually want around them.
The arrangement can be a relief at the company level too.
Being the author of every important layer means carrying every compatibility decision, setting every expectation, maintaining every weird corner, and receiving every complaint. A healthy ecosystem outside Microsoft gives the company places where it can participate while leaving many of the rules to somebody else.
Competition can even be useful in that sense. Somebody else develops the standard. Somebody else proves a product category. Somebody else has users who demand interoperability. Microsoft gets a thing it can follow instead of another domain where the entire industry waits for Microsoft to set the tone.
The claim stays narrow. Microsoft is a giant profit-seeking company and competes very hard where it sees value. It can also make a lot of money in a world containing strong independent ecosystems, and every thriving external ecosystem is one more place where Microsoft can sell something while sharing authorship of the broader experience.
Old Microsoft wanted developers writing Windows applications.
Modern Microsoft is perfectly happy when your Linux workload runs on Azure.
The burden gets shared. The revenue can still arrive.
Everybody gets mad at the landlord
Microsoft also occupies an unusually awful place in the blame chain.
Windows sits underneath whatever somebody was actually trying to do.
A user wants to print a document. Pair headphones. Play a game. Open an old accounting program. Join a meeting. Plug in a drawing tablet. Use a weird industrial device whose vendor last updated the driver during the first Obama administration.
Any failure along that path can become "Windows is being stupid."
Sometimes Windows is being stupid.
Other times an OEM shipped garbage, a driver is broken, an application made a bad assumption, an enterprise policy did something strange, or three vendors each own one piece of the failure. The user still experiences one computer sitting in front of them, and Windows is the name on the computer's behavior.
Microsoft earned this position. Windows became valuable partly because an absurd variety of hardware and software works on it. The company made billions from being the common layer underneath the PC world.
The privilege comes with a permanent emotional tax: everybody remembers the landlord when the sink leaks.
Apple gets to exercise much tighter control over the devices running its operating systems and can retire old ideas with a degree of authority Microsoft rarely enjoys on Windows. Google and Meta grew much of their core consumer experience as centrally operated services that can change on the server side. Amazon's giant businesses often meet customers as more distinct products.
Microsoft has a consumer operating system, an enterprise estate measured in decades, a cloud platform, developer products, business software, gaming, security, databases, identity, search, advertising, devices, and a long history of making those things interact.
Every interaction is another place where a reasonable local decision can produce an irritating global result.
So yes, Microsoft sometimes looks unfocused. It has an incredible amount to focus on.
A defense still gets to complain
Microsoft still earns plenty of criticism.
Users can reasonably hate account nags, unwanted product promotion, UI churn, broken updates, or an AI feature showing up before it earns its space. Enterprise customers can complain about licensing. Developers can complain about product direction. Competitors can complain when Microsoft uses an established position aggressively.
A defense of Microsoft gets more interesting when the criticism stays intact.
The useful claim is that a surprising amount of Microsoft's weird behavior can emerge from competent people facing a genuinely strange inheritance.
Every successful layer leaves a constituency behind. Every new feature can become a future compatibility promise. Every simplification threatens an old dependency. Every new platform transition creates pressure to move before the evidence feels complete. Every increase in capability gives the company enough confidence to attempt another frontier.
Then millions of people encounter the combined result through the machine they use every day.
Of course the company feels slow and manic at once.
One hand is hovering over a twenty-year-old compatibility switch, asking who still depends on it.
The other is trying to get the company into the next computing era before a rival owns the default.
Both hands belong to Microsoft.
Its old victories are still running.
It keeps trying to win again.