The Mythical Machine Month

When I started my first internship, I had a boss that insisted that The Mythical Man Month was one of the most important pieces of software literature ever written. I took his advice and ordered the book.

I eventually read a few chapters, lost interest in the book, and eventually gave it away to charity. It never really connected with me - on the first read. Looking back, it made a lot of sense. The Mythical Man Month was born out of experiences at IBM. I was just starting off my career at a small Windows shop in Alabama.

When we got stuck at the internship company, we called in consultants - with really specific specialization to come help us out. I remember being awestruck by the amount of money they charged per hours; It paid to be a knowledge expert in VBScript! TMM commentated more on larger companies tendencies - at the time - to hire more talent to solve a problem.

As I worked on more products, over the years, I began to see how the book had influenced the ways we planned and built software. The ideas felt like explicit references - "oh, we follow this principle" - and more like assumed ideas of how to build. I was thinking about the book the other day, when I realized that so many of its ideas have effectively been upended now.

But what - exactly - is replacing it?

I think we're in the exciting and terrifying moment of figuring that out. So many folks have figured out how to ship code with AI tooling. They're now testing out the ideas of AI practices. But the question of "what actually works?" is still very much in the air?

I'm wondering if the world feels more like The Mythical Machine Month now.

We've moved away from valuing specialized knowledge and more into a generalist ability to thread together a lot of different inputs - product, design, engineering practices. The question of what "good" looks like in this case differs drastically - depending on who you ask. I see it as a good and bad thing!

Its good in the sense that there are definitely numerous ways to be a good Software Engineer that utilizes AI practices. Definitions of good are often very culturally local, as a result. Sometimes it can even be on a team-by-team basis. Getting to define the rules can be fun!

There's a lot of think-pieces out there around how they’ve figured out how to ship with AI. Many of them can feel bespoke to the way that a company or product is run.

Claude Code is a really good example of this. They're a frontier model provider and they have certain advantages and constraints with their products. When you read this, maybe there are things that you can take away, but it largely reads as a really cool story for what could become one day.

Then I look at Stripe's minions product. I start to see something that's a bit more shaped to places I've worked at.

These pieces are almost like stages of abstraction that allow us to break down a concept or see if more clearly - depending on what direction you're looking at.

This brings us to the bad. Variance - at a certain range - can be incredibly hard to manage outcomes for.

While I believe that the playbooks for developing software with agents is pretty well established, the management part of things is very much still in development.

Software Management is ultimately interested in shipping products and keeping margins consistent - among a thousand other smaller things. These things tend to have more human and invisible elements to them.

That's what brings us back to the Mythical Man Month. It's more concerned with the human factors behind shipping software - rather than the programatic ones. The Mythical Machine Month would ask that same question! Its strangely not being asked right now and it could be the secret to establishing organizational structure that actually works for the AI age.

Updated on November 20, 2026