In the beginning there is architecture, well maybe


I was on a panel last year when someone asked that old chestnut:

“Surely before you start coding anything you need to design an architecture?”

As regular readers might guess I took a very simple stand:

“No, you need to get something that works, you need to show you can address the problem in a way that delivers business benefit. You can retrofit architecture when you have shown that your thing is useful and valuable. Sure spend some time thinking about design and architecture but this should be measured in hours not days. As the system grows return to these discussions but keep them to hours not days. Embrace emergent design and refactoring.”

Another panellist disagreed:


In 2018 you can tell someone is about to disagree with you because they don’t start “No” or “But” they start “And”

“Companies have databases, and data centres, and standards, and you need to spend some time thinking about them. You need to integrate to existing systems. Changing database schema’s, let alone database application servers is big work and you want to avoid it if you can.”

OK, I can’t remember his exact words but that is the tone I recall. I was told, I was wrong.

Despite all those very real concerns I stand by my position. OK, maybe I need to elaborate a bit so here goes…

The thing is, I have nothing against architecture. When you build a software system you will – before very long – have an architecture. It may not be a very good one but you will have one.

I encourage all developers to study software design and architecture. Coders make hundreds, even thousands, of decisions everyday when they are coding and there is an architectural angle to every decision. Coders need to be versed in good design practices so they make good decisions.

Nor do I have anything against guiding the architecture in such a way that it becomes better with time – indeed I think that is a necessity. Neither do I disagree with architectural conventions in a system or documentation, again they become more valuable with time.

What I’m not in favour of is: believing that you can set it from the start. I believe the more you try to predetermine the more time you will waste.

Architecture is nine-tenths what you have, and one-tenth what you think you have, or should have. Architecture exists in the minds and shared models of those who maintain the system as much as it does in the code.

There are some occasion when there is a completely blank sheet of paper to start with. New start-ups spring obviously, but even inside a large corporation there are occasions when it happens; say when the company wants to experiment with new technologies and approaches, or hedge options. There are two priorities here:

  • Be different, differentiate the solution from what already exists
  • Prove that this new solution delivers something of business benefit

So: pick the solutions that seem right to you now – thats “you” plural, you and the team you are working with.

Maybe spend a little time considering your options (Oracle? MySql? CouchBase? Mongo?) but don’t spend too long, hours not days. Accept that you will learn when you start using your chosen option and accept that you may change it, or that you may make a bad decision and be cursing you decision for ever more.

The thing is: all that time spent designing the architecture, researching options and making decisions is wasted if you do not deliver business benefit. If nobody uses it then having the perfect product, perfect architecture, is pointless.

Anyway, even if you spend months and agree the perfect architecture it will not survive. Once you start work you will find assumptions invalidated, you will find things don’t work as you expected, and you will find that the cheap code monkeys you hired to implement you wonderful design do things you didn’t expect.

Now consider the case where you start a new initiative in an existing environment. There are existing databases and systems to work with. Again you don’t need to spend a lot of time looking at options because you don’t have them. Instead you have constraints.

If the company standard is Oracle for databases you are going to use Oracle.
If you need to integrate with SalesForce then you need to integrate with SalesForce.

You might believe that CouchBase is a better solution than Oracle from the word go. In which case you can either:

  • Accept that right now Oracle is the standard and use it even if it is an inferior solution. In time if you demonstrate business benefit than you will be in a stronger position to change. Or you might find that even an inferior solution is good enough.


  • Argue your case and hope to win: if you don’t win then you are back to the previous option, if you do then great.

What I would not do is: embark on a lengthy examination of all the possible options with the aim of deciding which is best. This takes time and means you will deliver later, which means cost-of-delay will reduce the value you deliver. Sure the great design might generate slightly more business benefit, or get sightly more performance from your development team but you are also delaying.

If you start with a good-enough solution you will learn. That learning may change your opinion, or it might confirm your initial opinion.

Remember: You are not building the ideal system. You are building the best system you can within constraints. The best system you can with the time and money available. In an existing environment many of those constraints are usually given before you start.

Unfortunately, us development folk tend to limit options thinking and exploring to the initial stages of a development effort – usually before any code is written. Once work is underway they don’t consider options often enough, we tend to jump at the first solution we think of.

Like this post? – Like to receive these posts by e-mail?

Subscribe to my newsletter & receive a free eBook “Xanpan: Team Centric Agile Software Development”

Check out my latest books – Continuous Digital and Project Myopia – and the Project Myopia audio edition

%d bloggers like this: