2 ms·
Software Estimation mentions some interesting research on team size versus calendar months to implement a project of fixed size. The upshot is that a team of 2
by btilly 16y ago
Software Estimation mentions some interesting research on team size versus calendar months to implement a project of fixed size. The upshot is that a team of 2 does work faster than a single person. A team of 3-4 works a bit faster still. A team of 5-7 works slightly faster. And then teams go slower!
You get back to the same territory at around 20 developers, and after that productivity scales smoothly. (Albeit with horrible productivity per person.)
This real-world data confirms and quantifies the observation made decades ago in The Mythical Man-Month that as a flat team grows, the overhead for communication grows as the square of the team size, while the effort available grows only linearly. There are techniques to organize teams in a more structured way to avoid this pitfall. But the drawback is that the necessary procedures, documentation, etc significantly reduce individual productivity.
Oh, and to this add the fact that any time you hire more people you have to accept a temporary drop in productivity as that new person comes up to speed. I don't have a figure for programmers, but for general office work a month is standard. And I'm sure from anecdotal evidence that programming is worse. Much worse.
On top of this, as I'm sure you know, the more code debt you have, the slower it becomes to implement new features, and the more risk you have when you do so. Plus you're setting the foundation for an environment that the best programmers simply will not wish to work in. What is the cost of that?
There needs to be a balance. You can't try to create pristine code. But don't discount the costs of a poor foundation, and don't plan on being able to hire past its limitations. That way lies disaster.
- anonymouscto 16y agoAwesome answer, and very reassuring. Thanks! I'll target <= 6 for now, assuming I'm allowed by the team I hire to keep coding ;). I guess I'm less concerned with technical debt and more concerned with the product debt, which is a closely related concept I haven't heard discussed much. By product debt I mean partially-thought-through features rushed out the door that have been forgotten about. Even if the code is perfectly-factored, the product can come across to the end user as buggy because of product debt. I'm concerned that hiring hackers (even the very best hackers) will in fact exacerbate this problem, because we'll be able to push even more partial features out the door. At what point to you loop around in a startup and revisit/remove old features? Of course there are no hard and fast answers, and I appreciate the comment about balance. I'm trying to gather more perspectives and experience on this to better inform my position on this.