Building with Value: An Example

Egon Elbre
Jun 4, 2019 · 6 min read

Previously I described a process for developing software while trying to provide as much value and create understanding as early as possible.

How would one iteration of this process look like?

I’ll walk you through one iteration using the Go programming language while deliberately showing the steps more explicitly. In practice, many of these steps could be replaced by some external tools or libraries.

I’ll continue with the example from the previous post, an Issue Tracker.

The Simplified Process

  1. Structure Follows Value: find out what is the most valuable thing to implement right now.
  2. Spiking: we create a draft of 2–4 changes.
  3. Gradual Stiffening: build these changes incrementally,
  4. Cleanup: observe the wholeness and try to notice any small mistakes.

Structure Follows Value

“What provides most Value in an Issue Tracker that we can implement right now?”

We may notice the essential pieces from the language used.

  • I need to get an overview of the issues.
  • I need to find information about this issue.
  • I don’t have good visibility on issue progress.

We may also get this information from a domain expert. We also should try to understand why exactly it is important and how it fits into the rest of the system:

  • It shows what we are doing.
  • It shows what we have done.
  • It shows what we still need to do.

Issue is definitely a valuable part. Hence we must capture it in code.


To ensure that we can adequately represent the complexity of an Issue, we should create a package for it. That way, we make it significant and clear. At the same time, we create a locus of attention which allows to understand and examine the feature wholly and whether it is complete.

Of course, packages are not the only way to group things — there are also classes, functions, methods, constraints, etc. What you use to represent the idea will depend on how large, detailed, or important the contained thing is.

It is better to start with a notch larger grouping than is needed, and it is not difficult to make it smaller. However, the reverse, moving from smaller container to larger, is usually more complicated. e.g. Going from multiple packages to a single package vs. going from a single package to multiple.

We’ll add our first Spike to issue/info.go:

package issue

We must have an issue.ID to identify an issue uniquely. Each issue usually has an issue.Status associated with it. We need something to bring all the attributes together like issue.Info. Keep in mind that we are sketching the code and are not committed to this structure. We are not looking perfection, but rather a global view of how things will work together.

Notice that I don’t use long names such as IssueStatus, IssueInfo because the package for them already contains issue.

We also need some way to store and load those issues. The way we store and load them can change. Hence we should abstract this knowledge away. We create an interface issue.Manager for it. We put it into issue/manager.go:

type Manager interface {
Create(info Info) (ID, error)
Load(id ID) (Info, error)
Close(id ID) (error)
List() (issues []Info, error)

To get an overview of how we will use it, we write some usage code into main.go:

package main

It doesn’t matter in which order you create these pieces, and sometimes it is easier to create the usage code first, other times it is easier to create the implementation first. The thing that does matter is that both exist to ensure that we have the implementation details right and that we can integrate it with the rest of the code.

The usage code can also be sketched as a test; this depends on how the sketched code will be used, how it needs to integrate with the rest of the system and other factors.

Gradual Stiffening

Notice that we don’t have any runnable code yet, it’s fine, because until now we were trying to grasp what we are implementing and that all the pieces work together as intended.

Now we will step-by-step start to flesh out the actual structure, until we have a reliable and stable runnable code. We are in the beginning stages of our project, so there isn’t much to worry about. We should skim over our code and notice anything that doesn’t feel nice.

The first thing we may notice is issue.Created. What would info.Status == issue.Created mean? This suggests that we haven’t captured the intent as well as we should have. Let’s refine our sketch, info.Status == issue.Open sounds much better. Hence we change issue/info.go:

const (
Open Status = "Open"
Done = "Done"

In main.go the manager doesn’t feel clear; it feels like a fuzzy concept without specific meaning. There probably will be more things that need to “manage” things. Is there a better name for it?

What does the manager do? “It manages and tracks issues.” Here is a clue for a nicer name Tracker. We shall refine issue/manager.go into issue/tracker.go and change:

type Tracker interface {
Create(info Info) (ID, error)
Load(id ID) (Info, error)
Close(id ID) (error)
List() (issues []Info, error)

We also make all the necessary adjustments to main.go. At the end of this, we should have code that compiles however, it’s okay if it is not yet completely bug-free. We will do this in the next step, but gradual stiffening and cleanup are usually finished with a final cleanup pass.


Now we have a good idea about the feature and how to put it into code, we shall go over and fill in all the missing details and ensure that we have comments and a few tests and can use it in some form.

Here we could add an in-memory implementation of the tracker and then write some tests for it.

The other thing what we want to do here is to make it easier to understand and ensure whether the code behaves as it should. In most cases, you would wish to unit or behavior test, but they are not the only way. You could also write property tests. Or write output that could be verified by hand, if the correct behavior is difficult to describe in code.

Few interesting bits to pay attention to while solidifying code. When you come across questions, mark them as such. For example, while writing the tracker test case, I made a mistake while writing:

// ...
expect := Info{
ID: id,
Caption: "Caption",
Desc: "Desc",
Status: Closed, // compilation error, should be Done

I mixed up two things: the method is called Close , and the resulting status is Done. Because I made a mistake while writing this, it suggests to me that the code is not clear enough, but I’m not sure how to improve it. It probably isn’t that important, so I’ll mark it as a TODO and move on to other things:

const (
// TODO: should tracker.Create renamed to "Open"
Open = Status("Open") // Open means that issue is in progress
// TODO: should "Done" be renamed to "Closed"
Done = Status("Done") // Done means that issue is completed and delivered

I could try to figure out this immediately, but I really don’t think I have the necessary information right now and I will probably find out the details while implementing other things. In a similar vein, we might reconsider Tracker method names.

Now we have captured something of value in the code. There is plenty of things we can do to improve the ease of use that code, but we have something that someone would like to use already.

Using this process helps to keep focus on what matters and at the same time bring together a bigger picture.

Thanks to Taavi Kivisik for review feedback.

Egon Elbre

Written by

More From Medium

More from Egon Elbre

More from Egon Elbre

What is a Layer?


Also tagged Software Development

Also tagged Software Development

Testing Strategies for Front-End Developers

More from Egon Elbre

More from Egon Elbre

Manipulating JSON Fields

Aug 1, 2018 · 4 min read


Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade