Fred Brooks and the mythical man month observed that communication between people expands faster than linear, and that adding more people to a project makes it later.
> Fred Brooks and the mythical man month observed that communication between people expands faster than linear, and that adding more people to a project makes it later.
In my experience with more than a couple dozen software projects, the ideal team size to successfully deliver a non-trivial effort is between five to nine people.
One subject matter expert visionary
One technical architect visionary
One or two technical masters
One or two UI/UX masters
One or two padawans capable of becoming a master
Staffing exceeding the above for an individual team is more a managerial genitalia contest than anything focused on organizational success.
I wonder how much a good process of meetings, tickets or whatever sort of documents etc for handling new ideas and features would speed this up or bring it back to linear. I've been in some orgs where good organization at the level above me can make the required communication between me and other dev teams lower and let us live in "good" silos
I've heard that project architecture has a strong enough impact on feature completion rate that you can infer how tightly coupled vs loosely plugin based it is just from graphs of feature release.
I think it's kind of fun now. Agents can interact with Linear. I'm still playing around with like does an issue achieve anything? What is the fastest way to get a bug from the observing session to the originating session?
But it's secondary.
The wall is build. If you can AI program, your problem is the build is too slow, too unreliable, not secure enough from a supply chain standpoint.
That is where one competent senior hacker tops out today. The agents are yielding to a CI that gets 35% per-vCPU occupancy.
The wall right now is skill or build, depending on your skill.
What are you talking about? A properly prompted agent generates a project with a fast, reproducible, and isolatable build. One of the best things about agentic AI is that I spend far less time on devops and on waiting on builds.
There are low-insensity regimes where it's all Python or whatever, and if you're in one, great.
When you're dealing with multiple platforms, or hardware accelerators, or mostly all of economically relevant shit in the AI era you don't get a small, clean, fast build.
Fable can't print a Tauri faux-native app without dragging in half of LLVM.
For greenfield work, we are using mostly Rust, although small utilities built with Go or Python are also acceptable provided they have a comprehensive set of unit tests, an integration test suite, well-defined specifications and acceptance criteria, and thorough documentation. Once you have that, it's much easier for the AI to work on it. (Our main motivation for doing this is that it takes far fewer tokens now to get stuff done since we have all that, to the point I can leave Qwen 3.6 running in 24/7 loops accomplishing things.)
If your build is bloated and slow, it would seem prudent to go and fix that, possibly converting a codebase from some other language or build system to something that can be built and tested very quickly. Slow builds and slow unit tests are a choice.
I'll contend that any Rust build involving Cargo is bloated and slow. `rustc` is impressively slow on a translation unit basis, and Cargo is basically a build recursion bingo card. Throw in about 900 micro point releases in flight at any given time?
If you're not running an elite `bazel` or `buck2` RBE with `nativelink`? You're not even playing.
Amdahl's law says something similar, adding more processors to a problem eventually gives almost no speedup because parts of the problem cannot be parallelized and need to be done sequentially.
To the people wondering the meaning of the article, it's I think this.
amdahl's law is trivial tautological nonsense and has nothing to do with the mythical man month.
The mythical man month is about communication between people and the amount of work to be productive in a project. It's about people colliding with programming.
It has nothing to do with a vague statement that "eventually" you have diminishing returns on processors (of course there is some limit even though different tasks could be many orders of magnitude apart).
It's completely obvious and the "eventually" is where anything interesting happens. This comes from the early days of computers which is why something trivial became a supposed insight.
For Brooks the communication between people starts to dominate the time spent the more people you add, so it will eventually become the bottleneck.
Not just the communication between people while working in general but the communication to get someone involved in the project weighing down the people who would be making progress.
Feel free to come up with your own conclusions.
I did come up with own conclusion, the only similarity is the word bottleneck.
It might seem obvious if you don't think about it much, but one is a pragmatic breakdown from experience and one just says "adding cpus won't work forever if there is some part where you can't add more cpus".
One is basically 1 + 1 = 2 and the other is a warning of a non obvious situation that comes from lots of experience and effort.
In my experience with more than a couple dozen software projects, the ideal team size to successfully deliver a non-trivial effort is between five to nine people.
Staffing exceeding the above for an individual team is more a managerial genitalia contest than anything focused on organizational success.