18 July, 2008

In defence of the lone coder archetype

All too often these days we see companies and methodology evangelists touting the system du jour whether it be Extreme Programming (XP/Pair Programming), or Agile as being the "One True Path" to coding enlightenment. All other be damned is roughly what said mantra translates to in common speak. We're told to beware the engineer who works alone in a dark back room for weeks on end. We're told to such a thing is heresy against the gods of modern application development.
We're being lied to by the methodology evangelists. I'm not here to down speak any of the other methodologies currently in use for software development. I am here, however to point out the benefits of the 'lone engineer' (not to be confused with the lone gunman). Lets set a few things straight though.
A coder with little to no professional experience in the field is not the optimal individual for this kind of setup. This is more for a seasoned professional who has at least been a team lead on several projects from conception through maintenance phases. I'm not saying it isn't possible to pull it off without significant experience, but I wouldn't place bets on such an individual succeeding at the whole task at hand in said company.
This kind of environment requires a disciplined engineer. We're talking about someone who knows the questions to ask, and how to dig deep to find the true needs of the client(s). This also means that the engineer needs to stand his/her ground when it comes to setting a solid schedule for phases of the project(s). There is room for variability in terms of time frames for each phase, but that the plan must be laid out in a linear fashion. This is engineering after all, and one doesn't plan to build the rooms before the foundation is laid.
Agile methodologies are more akin (though not parallel to) modular housing construction. Individual sections or units can be build simultaneous and changed (to a certain extent) by separate sub-teams. Whereas the lone engineer mentality is more akin to traditional construction in which the design is finalised, the foundation is laid and construction occurs in layers from the bottom up with customisations being generally last.
This isn't to imply that software built in this manner cannot change or adapt in a timely manner, far from it. Seasoned engineers/developers understand that the only true constant is change. We constantly work on new ways in which to design flexibility and mutability into our systems regardless of the specifics. This however just goes back to one of my earlier points in that this requires someone who isn't wet behind their ears. You can only know and plan for the unexpected via flexible designs after having cut your teeth in real world projects, and by faltering. Everyone makes mistakes, and it is through these mistakes that we grow and better ourselves. I'm not implying that senior level engineers don't make mistakes, I'm just pointing out that their success to failure ratio is pretty strong in favour of the success side.
One of the key benefits of the lone engineer methodology is that focus can remain solid with no deviation due to intra-coder/intra-team conflicts. And while this can at teams be a negative, overall it proves itself beneficial towards the ultimate goal, a finished software package/eco-system designed from the ground up to benefit the requisitioner(s)/client(s).
The issues of coding standards still must be kept, however there is definite consistency when only one individual is involved. This is also a liability as said engineer must have strict standards, commentary and documentation from beginning to end so that in case of any situation rending the engineer on the project incapacitated in any form, another could come in and continue with minimal delay. These are issues which affect other methodologies, some more than others.
There are downsides as well though. There are no direct peers upon which one can depend in times of need or collaboration. There are no peers to assist in code review sessions, meaning that the QA of any project must be proficient at testing, and enforce strict recursion testing when any changes are made to a product in production as QA becomes that final line of defence against human error. The reality is, no usable (as in serves a real purpose) piece of software is bug free and to think otherwise is the sign of a fool. What is realistic is having a system robust to handle errors when they occur, even if that is infinitesimally infrequent.
I'm getting off track here. My point is that in this newer, more extreme era in which coding fundamentalists believe they possess the one-true-way of coding and all others be damned, we must step back and realise the grave miscalculation they are making so boldly. After all, those methodologies weren't what were utilised to get us where we are today, nor were they even thought of when some of the staples of our art were invented. This doesn't mean that the next great piece of software or language won't be produced in such a manner, as it/they may. It means that we as professionals shouldn't be so narrow minded with tunnel vision when approaching different methodologies in an attempt to find which works best for the project/company/industry at hand.