When is Enough, Enough?

Clayton Neff Programming 8 Comments

This summer, I had the pleasure of chatting with a distant cousin who was traveling through town. He is a retired project manager from NASA, but they keep bringing him back as a consultant, so obviously they like his work. I had never dealt with him as an adult, and he was insistent we talk when he learned I was a software developer. He asked me one of the strangest questions I think I have ever heard from a project manager:

“How can I get my developers to stop?”

Now, I’m sure we have all fallen into the trap of wanting to add just one more feature, tweak one more method, to be a bit better. But we also have deadlines, don’t we? And clients that we have to answer to?

It turned out that in my cousin’s case, there was too broad a gulf between the actual consumer of the software and the people developing it. They never spoke to one another. Everything was done through intermediaries. While I’m sure part of this was due to the nature of government work, where almost everything is contracted out (and you can’t have the contractors talking to each other without a government employee being involved), but this seemed more fundamental than that. It was what I thought was the extinct mix of waterfall and cowboy programming.

Calling something “waterfall” has become a severe term of derision in lots of shops these days. And while I think the waterfall process really should be derided, it is getting to the point where processes that aren’t at all “Waterfall” are being labeled as such because someone doesn’t like them. Unfortunately it is much easier to say, and more widely understood than “non-agile” or something similar. And while I don’t have enough information to know if these projects truly were waterfall or not, they were certainly not agile. And throwing an isolated developer into the mix only made the situation worse.

See Also:  More on Accessibility-First Programming

When discussing agile development techniques, the list of benefits is usually skewed to those for the developers. Less time spent in meetings and more time coding. Let the code be the ultimate requirement document. Things like that. What I think gets lost all too often (and far too frequently not utilized) is the engagement of the ultimate consumer (or their representative within the company).

At the end of each sprint, the developers should give the consumer’s stakeholder deployable code with something for them to test. If they sign off on it, it should go into production.

And here’s the important part: further development on that code should STOP. The stakeholder with the responsibility for making that call has said it is good enough. The developers should move on to other tasks in the backlog. If they have ideas for how the just released code could be improved, then that should be added to the backlog and prioritized along with everything else.

Of course, this requires that the stakeholder who is supposed to make the final go/no-go decision has all the relevant data. They can’t approve a new algorithm that hasn’t been stress tested, or a new UI that has only been tested in one environment. It is our job to present them all the facts, and then let them make the decision. And I have found that if you have been open and upfront about all the design choices and their trade-offs, and engaged them in the decision making process, they almost always make what developers would consider the “right” decision. However, we as developers must realize that even when they make the “wrong” decision after getting all the facts, it is our responsibility to implement what they ask for in the best way possible.

See Also:  Decoding Base64

Final Thoughts

The agile methodology and extreme programming techniques have given a lot of power back to the regular developers. We get to make a lot more choices about tools and architectural decisions about how things should be done than we did in the bad old (waterfall) days.

But we shouldn’t abuse that power either. As Uncle Ben said, “With great power comes great responsibility.”

We have to remain aware of who our ultimate customer is and keep them happy. If we keep missing deadlines, or don’t get some new feature in a sprint because we are tweaking something that is already good enough, more people will start to ask “How can I get my developers to stop?” And once they get you to stop, they may not ask you to start again, but ask someone else who knows when enough is enough instead.

— Clayton Neff, [email protected]

Comments 8

  1. This is where things like acceptance criteria and TDD can be really helpful. They define the expected checklist, and when that checklist is complete, you’re done.

    1. I am finding it really difficult to get much buy-in for TDD despite its many advantages. And I think we have to be wary of it making people complacent that they have defined all the requirements by defining the first round of tests. We still have to iterate through the design and implementation, adding and changing tests as needed.

      But you make an excellent point about how TDD can help define what “Done” is more clearly than many other methodologies.

      1. ATDD is a necessary counterpart, and is probably even more important than TDD. That becomes the meta checklist for delivering the feature. You and the client can write that list together, agree on it, and everything is directed to making those scenarios pass.

        That doesn’t mean the list can’t change, but it’ll only change through interaction with the client, and both parties have the same understanding at any given time of what “done” means.

  2. I whole-heatedly agree with this article. However, as business processes change, so do systems. I think the trick is trying to determine what tweaks make a material difference to the overall system and ultimately, support the business process. I find that changes will be requested based on an individual’s personal preference. This can create a vicious circle of changes for the sake of changing, not true improvent.

  3. I have also run in to the “vicious circle” affect, but I think that is a different problem. That is more a matter of two stakeholders not agreeing on a requirement, and the project manager not using the system to make them agree before creating a backlog item.

    When you don’t have a stakeholder that can truly represent the end user, you will always end up with “best guess” requirements. Sometimes we may not agree with that guess, but it is still our responsibility to code it as specified (perhaps in such a way that will be easy to change later).

  4. I had not heard of Gold plating either. I guess the idea is that we can always achieve some sort of perfect solution if we keep working at it. Thanks for the link.

What Do You Think?