relaxdiego (Mark Maglana's Technical Blog)

Automated Testing, Organizational Learning, and Thermostats

Jan 28, 2015
Est. read: 4 minutes

Agile software development is oftentimes touted as being a faster and more efficient way of delivering software, as a way to reduce time to market, and as a means to quickly respond to constantly changing requirements. That’s all well and good until the message is misunderstood to mean “let’s skip documentation and testing and just fix things as we go so that we can deliver within the arbitrarily set deadline.” That’s not to say that Agile is no good. Rather, it’s to point out that the need for software development speed must be tempered by the need to retain organizational learning.

Anatomy of a Dynamic System

Dynamic systems typically have an input, a controller, and an output. That is, you tell the system the desired output and it will perform a series of internal operations to make that happen. Modern dynamic systems, however, have a crucial component that makes it more effective: the feedback loop.

Source: http://en.wikipedia.org/wiki/Control_theory


The feedback loop captures information about the external environment and feeds that back to the system which uses that information to regulate itself. To put this in more concrete terms, let’s look at a heating system’s feedback mechanism: the thermostat.

Source: http://en.wikipedia.org/wiki/Thermostat


To raise the room temperature, you’d slide (1) to the right. This makes (2) turn clockwise which, in turn, moves (4) to the right. Notice that (4) is connected to the wire marked (3) which is one end of the circuit. Move (1) to the right just enough and (4) will make contact with (5) which is the other end of the circuit. This closes the circuit and fires up the heater.

The feedback loop in this case is the bimetallic strip that’s wound into a coil surrounding (2). Yes, in this case, the feedback loop is literally a loop! This coil is continually collecting information about the environment (specifically, the room’s temperature) and feeding it back to the system. When the room heats up, the coil unwinds, moving (4) slowly to the left. Uncoil enough and (4) will detach from (5) turning off the heater. As the room cools down, the coil winds back, moving (4) to the right until it touches (5) turning the heater back on.

What does this have to do with agile software development? Everything because software development processes are dynamic systems too: the customer’s requirements are the inputs, the software design and development activities are the processes, and the software is the output. It also has a feedback loop that contains more information than a thermostat would (user feedback, bugs, etc.)

Waterfall Revisited

A typical waterfall process is commonly understood as having the following phases:

Source: http://en.wikipedia.org/wiki/Waterfall_model


What is clearly absent from the waterfall model is the critical feedback loop that would have allowed it to respond to new knowledge learned along the way. Waterfall software development is like a heating system without a thermostat. It’s bound to burn the team and the customer in the end.

To be somewhat fair to the waterfall methodology, however, a feedback loop is possible but it tends to be so late in the process (verification phase) that it’s virtually worthless.

“Agile” Software Development

Agile’s promise is to deliver correct solutions sooner by essentially slicing major phases of the waterfall methodology and iterating through them in sprints. When you think about it, these sprints are really mini-waterfalls with the demos at the end serving as the output and also the source of the feedback loop. The advantage here is that because the feedback loop happens sooner and more often, it is more actionable and useful. However, if the knowledge learned in each sprint is not codified in any way, then that feedback loop is anemic at best because people forget, team members get re-assigned, employees move on, and organizational knowledge is lost.

You’ll often hear or directly experience these kinds of “agile” team environments where they start out delivering software very quickly but then get mired by difficult bugs or an inflexible architecture. These types of teams often hold the banner of agile but skimp on automated tests and some form of documentation to save time.

Agile Software Development

True agile teams take full advantage of the feedback loop that the process offers. One of the ways that they do this is by codifying knowedge via automated tests that remain relevant throughout the project. It is through these activities that a team is able to sustain its velocity and success. If you find that your team is unable to consistently deliver, check your feedback loop. Chances are it’s not been given the respect and attention that it needs.