▲ up

Don't fret over it not being perfect

Software is such a young and unique stream of engineering. Almost always we cannot physically see the product that we produce as opposed to civil engineers for example, who have their buildings and bridges to see and test. Along with this however comes a much higher cost and risk of their design requiring to be right the first time round. On top of this many people will be using these final products and a failure won't mean that they can't double tap to like some celebrities new fad diet.

Whilst this may sound like it is belittling the work we do as an industry, but the truth is this is a super power that we have over every other stream of engineering. Once they have built and shipped a car to the end customer, they cannot simply add a minor improvement to the brake pads and then update all existing customers and still make a profit. We however can! This iterative development cycle is one of the key factors of agile development and our ability to adapt to the change in requirements. This is why we can pull out concepts like Kanban and Scrum, and then single out the deficiencies of waterfall in our industry, something that works so well in others.

In my opinion one of the core pieces and that which I wanted to share some thoughts on is the iterative versus upfront design. In particular my experience with this in a team working with Kanban and how important it is reflect on your working process to avoid eventually grinding to a glacial speed.

Iterative vs upfront design

One of the key aspects of Agile is in the name itself, have the agility to change how your product develops. This is as opposed to the classic approach of completing all the design upfront before we even build the first tiny piece. In this upfront design we will typically be inundated with documentation and discussions on all possibilities. Again I emphasize that this has worked well in numerous other industries but in the software industry I just don't believe it is a good fit. Doing all this design upfront forces us to ask all the questions and have all the answers before we can even start to test any of our hypotheses.

I have found that we can also use this mindset from the scale of an entire product, on a single service, or even a single card from the backlog. This single card level is where I want to focus on for the rest of this post.

Iterative development with Kanban cards

Previously I had worked in a team following the Kanban approach, I won't go into details about this style of working here, but the main point is we had our projects sliced up into smaller tasks which we would pick up on to work on. It is these tasks or cards that I started to notice we were falling into a lot of upfront design, gold plating, and what if scenarios. At first I thought this to be good practice, designing for the worst and hoping for the best. It wasn't until after a few months that I noticed this was starting to decay our work process, card scopes began to increase once work had started, cards were starting to overlap with other cards making dependencies we didn't want, as well when it came time to hand over we had to define a whole new set of definition of done.

My thoughts as to why this occurred was based on the fact we were prematurely gold plating our solution, particularly since it was a new project were we were continually discovering new requirements. My learning from this can be boiled down to the following topics:

Watch the scope creep in cards

In our work process we had groomed all the cards prior to them being ready to be picked up, that meant there was a pretty scope of what was required from the card, which had input from the developers. However there will always come a time when you miss a subtle requirement or you find to do this you need that. At this point it is very tempting to switch gears and take on the new found task but you should avoid this as it then introduces new scope into your current card. Particularly in a Kanban style of working where you are focussed on flow of work, increasing the scope of a card is particularly problematic.

When you do find these new requirements or pre-requisites, don't view it as a fault in your process for not knowing this beforehand but rather see if you can carve up a new card to capture these new requirements. Doing it this way means that any dependencies on your current work won't get blocked and it may even be able to be picked up by another team member immediately so it can be completed in parallel.

If this new piece of work is a dependency on your current card it is still possible to split it into a new card and not block yourself. Consider how you could put in a stub or something of the sort which will allow you to progress and for that stub to be removed/implemented as part of the card you have just created.

Define and stick to an acceptance criteria

One of the great parts of the work process I had in this team was that we had a close working relationship with QA. When we picked up a new card we defined an acceptance criteria or a definition of done for that card. This was typically a bullet point list of some use cases QA could test against and what they were to expect. What I ended up learning was that this was as useful to me as a developer (if not more) as it was for QA. If I stuck to only implementing this acceptance criteria then I could easily avoid over complicating the implementation. This meant that it was even more important that I participate and understand where this acceptance criteria came from and really focus on breaking down that barrier that we see all too often between developers and QA.

Only focusing on the acceptance criteria can mean that you know that there may be some implementation detail you believe is missed. Again write this up into a new card because time and time again I have found that I expected something was required only to find out later it is covered by something else, requirements change, or I made incorrect assumptions. It more than likely is better to write a card which captures your thoughts in the moment and reevaluate its importance later on when you are ready to pick it up.

Utilize user stories to understand what you are achieving

Finally when implementing a new piece of functionality it is always so much more satisfying if you know the impact you will be having on customers. This is where user stories provide the answer. Defining at least one user story for the card you are going to pick up feeds into a lot of positive processes indirectly. Having a user story means that defining acceptance criteria will be simpler as it is based on this user story. When you are considering the scope of work you need to complete again this user story provides that basis on which you can check if your expected implementation will handle that scenario.

Whilst I have mentioned the user story to a particular card sometimes this is difficult and one user story may span several cards and hence the right place to create such a story is during the grooming session.

The main aspect of gold plating this user story achieves is it helps us bring our attention to what actual functionality we want to deliver to our customer and not all the ones that we think we want to deliver.