From 2004 - 2014, an pseudonymous employee called Mini-Microsoft frequently posted critiques of Microsoft management from the perspective of a technically-literate manager in a series of hot takes. Top hits include:

- Vista 2007. Fire the Leadership Now! I echoed this sentiment more extensively in my Broken Windows Theory post, for which I nearly was fired by Jim Allchin. (You’ll have to do some serious internet sleuthing if you’re actually interested in resurfacing a publication which I’m not terribly proud of).

- Microsoft Annual Review 2011. The post itself says little, but the comments were like Blind before that was a thing.

His pseudonym was inspired by the leitmotif that Microsoft had become too big and bureaucratic, and that it’d be much faster if it was reduced to half its size. (Remember that in 2004, Microsoft was 57,000 employees, whereas it’s now 223,000 — so I can only imagine Mini’s will has been absolutely pulverized).

The irony is he’ll likely soon be right, decades after issuing his clarion call to go lean.

Fireside True Story™ Time: Over the years, many people have asked whether I’m Mini-Microsoft. I’ve denied this time and time again using what I feel to be pretty simple evidence:

First, Mini shared much information that suggested he was a Partner. (And FWIW, it’s known he’s a he). I was never privy to partner meetings. Second, our writing styles are completely different: not only the words we use, but Mini’s tone is far more cynical / negative / critical. Third, I left Microsoft in 2010, years before Mini stopped posting.

The Firm

Three years ago when AI first became mainstream, there was a lot of talk about whether we’d “do more with more” — that is, whether the increased productivity of each employee would motivate companies to simply grow their ambitions. This line was especially popular with CEOs, since journalists harbored a suspicion they’d actually do more with less, trimming the ranks of employees to recognize record profits.

The argument for doing more, according to CEOs, sounded deceptively simple: “Every team in my company would love to have more staff. If each employee suddenly produced more output, we obviously already have the ambitions to absorb that.”

Here’s the problem: every top tech company already has profits far exceeding what it’d cost to explode out their staff. Microsoft made $134B profit in FY2026, when they had 223k employees each costing ~$275k (including benefits, offices, IT, etc). That’s enough to triple their workforce using profits alone.

But they weren’t tripling their staff, even before AI. Why would they do it now?

There are at least two reasons why Microsoft didn’t just blow out its staff to 670k employees using its fountains of cash:

- It’s impossible to upkeep the quality of each hire when you reach those quantities. Amazon has a “Bar Raiser” philosophy, which is great on paper if you have a company of, say, 3 people. But when Amazon already has 1.6M employees, you’re going to tell me that you’ll get to 1.7M by hiring 100,000 people who all are above your current average employees’ abilities? Regression toward the mean kicks in several orders of magnitude before that.

- Coordination efficiencies kill your per-capita productivity. Mythical man month is one classic argument against adding more people. We’ve all seen cases where adding an additional member to a team reduces total team output. But there’s a far more important reason which has to do with the “theory of the firm.”

Anyone who’s been to Microsoft’s headquarters knows that The PRO Club, a ritzy fitness megaplex, was built solely on Microsoft benefits dollars pouring one golden brick at a time into their well-toned coffers. Every employee has gotten a sponsored gym membership there for nearly five decades.

In my early twenties, seeing how ridiculous of prices The PRO Club was charging Microsoft, I used to wonder why Microsoft didn’t just operate its own gym. I don’t ask that anymore, having been taught the Theory of the Firm.

The central question is, “Why are companies the size they are? No bigger, no smaller?”

In 1937, Ronald Coase proposed the answer is “transaction costs.” If transacting were easy, your employees would focus like a laser on the one thing you do well and outsource every other function (e.g. HR, payroll, legal, recruiting, facilities, sales); a crack team of engineers would outsource PM, Design, and Sales.

But why isn’t this done? Can you imagine how painful it’d be to negotiate and manage all those contracts, especially when something goes wrong? Every function you bring in-house turns an endless series of contracts into one contract: the employment contract.

If bringing functions in-house reduces overhead, why does Microsoft pay The PRO Club, a lawn maintenance company, a physical security company, and also a janitorial service? The reason is that corporations have internal costs which scale with their complexity. “Mythical man month” is exactly that — the idea that communication and coordination overheads can begin to exceed the work itself.

Here’s Coase’s answer to why companies grow to a certain size and then stop:

A firm will tend to expand until the costs of organizing an extra transaction within the firm become equal to the costs of carrying out the same transaction by means of an exchange on the open market.

Whither Next?

We’ve explored why Microsoft doesn’t just hire 3x more employees. Let’s turn our attention to what any large firm should do now that AI is making employees more productive.

You could of course stay the same size you are now. For a short while, this’ll lead to seemingly higher productivity as each employee produces more. But this temporary bliss won’t last.

- The biggest efficiency lever in any sufficiently large organization is to reduce staff — not because of the payroll savings, but because of the reduction in coordination overhead. Anyone who’s worked in a company of tens of thousands of employees knows this is true. Would you rather have a team of 10 people each producing a widget, or hire one person who can create 10 widgets?

- Coordination overhead is about to go through the roof. People at Meta report this already: when you make a team 4x more productive, people start colliding with each other all the time. Merge conflicts. Misunderstandings. Two different implementations of the same thing. Meetings to sync on next actions. Can you imagine a daily standup where each person has 4x more stuff to report on, 4x more stuff which could potentially require talking to another engineer to avoid butting heads? Jamming 10x cars onto existing highways doesn’t mean 10x more people transit; in fact, it means everybody gets to their destinations more slowly.

- Interfaces are about to get much cleaner. One of the major transaction costs Coase identified was the difficulty of negotiating and enforcing contracts between organizations. API keys and transaction-priced MCPs are enabling companies to establish collaborations via low-overhead self-service. Take Figma. You used to have to schedule meetings with your designer, ask them to export various resolution screenshots for your App Store submission, etc. Now you just connect to Figma via MCP and implement the design directly. We used to talk about microservices architectures within companies; we’ll soon see an explosion of companies offering services through API and MCPs requiring no contract negotiations and objectively-specified deliverables.

All of this points to shrinking corporation sizes. CEOs can’t easily say this, of course, just like Uber had to talk delicately about their autonomous driving vision for years before their “partners” were inured to impending reality.

This shrinkage, by the way, doesn’t only apply to corporations focused primarily on intellectual outputs. As robots improve, the same logic will apply to organizations with large physical footprints.

What About Me?

This isn’t necessarily bad news. In fact, this could be great news for you and your job.

First, corporations becoming much smaller doesn’t necessarily mean there are less jobs. The same people who work today can be reconfigured into smaller clusters without anyone losing jobs. In fact, a burst of entrepreneurship empowered by AI could create a bevy of small companies each hiring employees away from large firms.

Second, stock-based compensation of executives could incentivize them to increase their firm’s size despite everything mentioned above. A partnership of lawyers or surgeons rarely votes to expand their partnership by adding someone who brings in far less revenue because doing so dilutes each partner’s profit. That sort of arrangement encourages maximization of per-employee profitability. But large corporations’ stock incentives don’t work this way. A CEO is incentivized to increase absolute profits even at the cost of reducing per-employee productivity.

Lastly, a Cambrian explosion of smaller firms could be a boon to two types of employees: people who’ve always wanted to start their own company, and people who are tired of all the corporate bureaucracy they deal with every day. If you fall into one or both of these categories, the coming world could be an amazing thing.

If I were a betting man, I’d say the preponderance of reasoning suggests we should expect average company size to shrink significantly as AI increases employee productivity. July 2026 was already the highest single month of new business applications in American history (576,512). We’re at the beginning of a revolution in corporate configuration.