Agileista

In a development environment it can be tricky to know what to measure. There are plenty of potential performance metrics to choose from, everything is done on a computer so naturally you can track and measure it. The problem is one of choice. Whatever metrics you define as your criteria for success will become the focus for improvement. Hence the power of both measuring and rewarding salespeople on pure sales numbers.

For developers, measure lines of code (LOC) committed and they will duly write more LOC. LOC alone means nothing; it could be crappy code right? It's better to measure for the outcome you want. Find some way of measuring quality and that becomes the incentive. Measure it and the team will naturally seek to improve it.

In Scrum there is the team's sprint burndown, which shows how quickly they get through their story points and how likely they are to finish the sprint tasks. A longer trend you could measure is release planning, where you do a similar thing to a burndown but for the whole project/release.

This is all fine for estimating and project planning but in measuring quality it's not much help. Again, what you measure as 'quality' will depend on your situation but some possibilities include: unit test coverage, defect count, uptime, customer feedback, use statistics. Whatever it is you measure, make sure you can get the data reliably and at the right level of detail.

Like it or not, visibility is one of the traits that Agile brings to the party. This visibility is a natural output of Agile process and forms an important part of the inspect and adapt feedback loop. You must be able to see what's going on so you can diagnose problems and make appropriate changes.

What you see of your project depends on what you measure; what you measure will be affected by the measuring. This is an important and related feedback cycle - setting metrics will shape project outcomes (I'll cover this in detail in my next post)

One of the metrics collected in Scrum is the estimated number of task hours remaining in the Sprint. We can use this information to create what is called a Burndown chart. A burndown chart is basically the team's remaining estimated time plotted against remaining time. The shape of this chart can help the Scrum master to see how well an iteration is going and to plan ahead for the coming iteration. To keep track of the software build as a whole a Release and Product Burndown can be used. This will track team output vs. time to release

The first step is in estimating tasks for users stories. User stories that have been chosen for a Sprint need to be unbundled into smaller actionable tasks. By doing this the team get a much clearer picture of the user story and tasks can then be estimated in 'man hours to complete'.

Estimates for each active task are updated by team members as they work their way through a Sprint. When a task is complete it is updated to show zero hours remaining. If a task is not completed by the end of that day it is re-estimated to reflect time remaining to complete it. This information is used to track the amount of work hours remaining in a given iteration; the Burndown chart.

This detailed level of time tracking can make some developers feel uncomfortable; they can feel that they are being monitored for efficiency like workers on a widget production line. When this happens team members may not update their hours on tasks and may slip into simply reporting a task is finished - some may not even do that!

Sprint Burndown charts become inaccurate and difficult to create when this happens. The teams ability to work together can be inhibited by this kind of scenario too. No one knows what's being worked on and the inevitable clashes begin. Luckily for us Boris Gloger and Tobias Mayer have come up with a simple and elegant solution.

During Sprint planning the team should not estimate the amount of time to complete a task, but simply ensure that each task is sufficiently small enough to be completed in a work day. By doing this a Burndown chart can still be produced and the team don't feel like they are being production line managed.

A by-product of this approach is that the team task board becomes much simpler and more effective. Every day during the 15 min Scrum meeting team members move their chosen tasks for the day into the In Progress column of the board and throughout the day those tasks are completed and moved to the appropriate column. This provides the transparency and team visibility that Agile promises and management demands.

The idea is to define and agree a basic way of working that allows us to focus on writing great software that satisfies the business. By using some good engineering practices we remove the waste in a developers day and allow them to get on with coding.

What follows is a mixture of engineering practice and working practices that I've had some success with.

1. Control your code - It's just good practice to use revision control however you decide to run the project. We like subversion.

2. TDD/BDD - Test Driven Development or Behaviour Driven Development. Write your tests first and then write your code to fulfil those tests. This way you don't end up with no time to test and it means you're doing the design up front in small chunks. People often wonder where the Architecture is in Agile, well if you do TDD the way it was meant to be done then when you write your tests you're actually designing how the code should behave. Effectively the TDD cycle of test-code-refactor is an inspect and adapt design process, that a developer goes through in cycles lasting minutes.

3. War room - Your team and their interactions are the single the most important factor in the success or failure of a project. If you can get all your team in one room and make it their war room it will help the team to be cohesive and develop a sense of shared values and team practices or behaviours.

4. Continuous Integration and Automated Testing - Automate both unit testing and acceptance or functional testing. Ad hoc testing is a bad idea - by automating the building and testing in the code control system you ensure that the software builds at every check-in; this is continuous integration. Developers will feel more comfortable changing existing code and refactoring, no one forgets to run the tests, you spot errors early, development becomes more flexible. For a test framework in Rails apps there is Rake, Zentest and Test::Unit for Ruby, and for JUnit for Java. For acceptance testing there is Selenium and Fitnesse. Cruise control for Ant enforces continuous integration in Java.

5. Quiet time - When you're in the zone and cranking out quality code like it was gong out of fashion, the last thing you need is an interruption. Once you've been rudely awakened it can take ages to get back into the right frame of mind for problem solving. Agree a quiet time with no phones and email or devise a physical signalling system like a busy flag so people know not to interrupt.

6. Automate those documents - Use software that extracts relevant information from the code and creates some of your documentation for you. That way your documents show you what is really happening with the software. It almost documents itself. So much more useful than some spec that says what should be happening.

You should also look at pair programming. Once I've had more experience of it I'll post my thoughts on it. Check Ron Jeffries site for more.

Remember, not every practice works for every situation, try them out and keep what fits. Make sure to get the team to agree their working practices - enforced ways of working that validate and measure the wrong practices and metrics wont deliver you software, let alone agility.