It seems these days there are countless methodologies, processes and practices to choose from when developing software and, somewhat ironically, the list seems to be growing at the rate of Moore’s law. I’ve read about, discussed, been on courses and been to conferences about a lot of them and the thing I’ve consistently found most useful is talking to other practitioners about what they’re doing and what’s working (or not working) for them.
Recently there’s been a lot of (and subsequently criticism of) debate on message boards and blogs about the relative benefits of one paradigm versus the other. Personally I don’t care much for subscribing to any particular paradigm and am much more interested in what works and what doesn’t (and in which circumstances) so my response is to publish what my team and company is actually doing right now. This is a snapshot of our current practices. Ask me again in 6 months and hopefully I’ll show you something very different.
The inspiration and influences for the way we work mostly come from Agile, Scrum, Extreme Programming, Software Craftsmanship, Lean, Lean Software Development, Kanban Software Development and The Toyota Way. It is all and none of these things.
Context
We are a small to medium size “start up” organisation working in the new media industry. The company employs around 60 people mainly based in the UK. The development department numbers around 20 co-located people. Agile practices are a relatively new introduction and the previous approach was of the typically chaotic type familiar to young businesses. We mainly work in .Net using C# but also dabble in Ruby, Javascript and UI languages. The rest of the article mostly relates specifically to one of the teams working within the department but also addresses some of the practices of the department as a whole.
Team
The team is made up of 4-5 developers and a tester. There is no project manager at the team level – in the spirit of self-organisation (principle 11) the duties traditionally the responsibility of a PM are shared between the team members. The Product Owner role is shared between the stakeholders within the organisation for the products the team is responsible for.
Iterations
We currently work in 1 week Iterations. It’s a new team who are also new to many of the agile concepts and doing this enables us to control the amount of work in progress, focus on delivery, improve our discipline and most importantly have short feedback cycles so improvements can be discussed and applied frequently. The downside is the overhead created by the amount of meetings. Once we’re comfortable the team is working well together we will have the opportunity to change this if desired (e.g. Continuous Flow, changing the iteration length, changing the frequency of meetings).
Meetings
Each iteration we have the following meetings:
Work Prioritisation – occurs iteration minus one. Stakeholders come together to raise and prioritise work not yet commited to.
Requirements Gathering – occurs ad hoc when necessary. All the team is required to attend along with the customer/s to bash out requirements for work prioritised in the prioritisation meeting.
Planning – occurs at the beggining of the iteration. Prioritised features (MMFs) which have been analysed are broken down into stories, discussed, estimated and commited to based on our current velocity (avg. over last 6 weeks)
Stand Up – occurs daily at 10am at the task board. Anyone outside the team is welcome to watch
Retrospective – occurs at the end of the iteration. Any actions from the meeting are to be completed by the end of the next iteration.
Requirements
Features are requested at the prioritisation session and use the User Story format.
More detailed requirements are gathered during the requirements meetings mentioned above, with the customer/s and all team members present. We use whiteboards to bash out the requirements and convert them into acceptance criteria using the “Given, When, Then” format. We have a rule that no work can be commited to unless we’re happy we have a clear understanding of the requirements.
Task Board
The task board is essentially a Kanban board with each stage of the delivery process separated into columns. We have an implicit limit of 2 stories in active, but otherwise have not applied limits to any other columns. Features (MMF) are blue, the stories which make up the MMFs are yellow, bugs are pink and quick support tasks white. When a story is commited to, the feature card is moved into “commited”, above the titles of the columns and tracks the last related story.
Measurement & Metrics
We use an Excel spreadsheet to hold the product backlog and track the data from the Kanban/Task board. Whenever an MMF moves to another column the date this occured is recorded. You can download a copy of the spreadsheet here (you may want to check the calcs on the CFD, not sure they’re right). Among other things it calculates average cycle time, average velocity and projections based on velocity. I’ve tried a few bespoke tracking tools (such as Mingle) and found nothing is as powerful and flexible as Excel.
We have a manual Cumulative Flow Diagram (CFD) which each team member takes turns to update daily so everyone shares the responsibility (it is also their job to update the Excel spreadsheet each day). The CFD diagram only tracks the value delivered to the business (one unit = one MMF. Measure the output, not the input) and is also represented in the Excel spreadsheet. Why have both you may ask? Visibility.
We have some rudamentry code metrics set up through our continuous integration framework such as NDepend output and test coverage but are working towards something more visible and useful.
Estimation
Still very much a necessary evil.
For comitting to work for an iteration we use Story Points using the fibernache (1, 2, 3, 5, 8…) sequence and achieve them by playing Planning Poker with everyone who may be involved in the work required to take part. We will only estimate (and commit to) work we have already analysed and gathered requirements for.
For longer term planning, as we don’t yet have enough information to be able to use cycle time for projecting work completion, using the velocity based on points completed per iteration has proven a very powerful toool to be able to give the rest of the business a better idea of our capacity and timescales (previously they had none). However this has well known drawbacks and we must be careful it does not get abused, as I have seen before (such as gaming of estimates, whether intentionally on subconsciously). Also, as we need to understand and have gathered the requirements to be able estimate this way it means there is very limited scope to how far into the future we can do this with any degree of confidence (as requirements will change). Once we have a reasonable amount of data in the system we will be able to use average cycle time, which will be much more powerful.
Coding Practices
Apart from the rules we’ve commited to as a team, Pair Programming, Unit Testing, Refactoring and the best working principles and practices of the software industry are encouraged from the top of the department and applied rigorously but pragmatically.
At the request of the department members (as a result of a disucssion on collective responsibility) we created a development standards document which includes topics such as naming conventions and testing. As much as possible the document is vague on implementation details to prevent it from holding us back when better working practices come along. We use shared Resharper and Visual Studio settings to help us keep to these standards.
As mentioned below we also frequently hold sessions to improve our skills.
Automated Testing
All new or modified code is covered by unit tests, integration tests (such as database interaction) and automated acceptance tests which test against the acceptance criteria (this last one is quite new territory).
Continuous Integration and Deployment
All projects are under continuous integration (we use TeamCity) and we are working towards having all deployments doing the same. We have monitors on the wall which show all the currently failing builds. Do I need to mention we use source control?
Roles and Responsibilities
Every role in the department is covered by a document explaining their roles and responsibilities. They are written in a way which encourages self-organisation and collective responsiblity. You can download them here. I will be talking about these descriptions more in a future article.
Learning Culture
Each week two hours are set aside for learning sessions such as coding dojos and presentations (we’re currently running a design patterns study group). Outside of these developers are actively encouraged to take the time to learn new practices during working hours (within reason). We have a library of books available on a range of subjects which are at the disposal of everyone. More often than not if there’s a book that someone would like to read the company will purchase it and add it to the library (books are pretty cheap in the grand scheme of things).
Continuous Improvement
Outside of retrospectives we have a monthly departmental session where the most pressing problems are discussed and actions taken away. However there is no limit or retriction to when improvements can be made and everyone is encouraged to take the initiative when they see a problem that needs addressing.
I was throwing some shapes on Twitter recently about some concerns I have with the current Kanban craze. Unfortunately I think the cursed 140 character limit meant my points got misinterpreted and may have lead people to think I’m anti-Kanban which is not the case in fact it’s quite the opposite. I’ve been using Kanban boards for over a year and half and jointly ran a presentation at XPDay2008 on evolving from Scrum to Lean which focused heavily on the use of Kanban boards.
The thing that’s making me itchy is how Kanban has somehow been elevated into a methodology unto itself. We don’t have “Scrum and Sprint” conferences or XPandPairProgrammingDay so why do we have Lean Kanban Conference Miami or the UK Lean and Kanban Conference? Also, pretty much everywhere you see someone talking about Lean software development the title of the blog or presentation also includes Kanban in the same breath? More than that I see a lot of discussion around Kanban in blog posts and Twitter but very little on Lean or the Lean Software Development principles.
I’m sure proponents of Kanban will say no one is suggesting Kanban is a methodology and I would agree I’ve not seen anyone say it is. The problem is interpretation. People have a habit of focusing on rules and methodologies because they’re a lot more easy to tackle than the problems they we’re created to solve. Scrum has been enormously successful (if you consider wide adoption a measure of success) but very few teams are doing it well as James Shore has been writing eloquently about recently because it does not force you address the real issues. The beauty of Lean software development is it is just a set of principles. It intentionally avoids prescribing how to do something. Obviously this causes problems as most people don’t want to get involved in the difficult stuff they just want to be told how to do it. Consider this reply I received on Twitter:
erwilke @robbowley “I think we’re trying to avoid kanban being seen as a stand-alone methodology, but people don’t “get” it as a set of tools”.
Maybe there’s a good reason why people don’t get it – because it you need to understand where it’s coming from. Focusing on Kanban and ignoring all the rest however, that’s easy!
Elevating Kanban to the prominent position it is now in makes me feel like history is going to repeat itself. I prophesied this some months back. It has been the most popular post on my blog by a long way.
If you’re getting into Kanban, be warned. Kanban is just a tool and in my opinion no more important than say, pair programming, unit testing or domain driven development. It is certainly a lot less important than the white elephant in the room which very few people seem to be addressing which is building the right thing in the first place. As Peter Drucker famously said: “There is nothing so useless as doing efficiently that which should not be done at all”.
Kanban is a small part of something much, much bigger, see the whole.
*Edit* Some responses to this article:
Is Kanban Just a Tool? - David Anderson
It is Not What It is that Really Matters - Israel Gat
Kanban: It’s a Tool, and There’s No Such Thing as “Just” a Tool - Project Management Revolutionary
Recently, whilst building a very complex work flow system for my organisation we joked how it would be much less expensive and more appropriate if we just gave them a few whiteboards and a stack of coloured index cards instead. It was working really well for us so why couldn’t it work for them?
I’ve just finished reading a few articles from my current favourite blog Evolving Excellence on the phenomenon that companies seem obsessed with trying to automate every possible process and are happy to spend gazzillions doing so, when a simple kanban would be far more effective and a darn sight cheaper.
Quote: “‘Excellence through simplicity.’ To me that quote from Lao Tzu has always epitomized one of the fundamental tenets of real lean. Don’t proceduralize complexity, and don’t make something more complex than it needs to be.”
http://www.evolvingexcellence.com/blog/2006/06/forget_sap_run_.html
http://www.evolvingexcellence.com/blog/2006/04/keep_one_eye_on.html
The workflow system we delivered is barely used by the customer (who asked for it) because it’s too rigid and every possible scenario is not accounted for. They’re very keen to make improvements (which will cost money), so maybe I really will suggest a kanban board instead this time.
It’s startling to see how much momentum has been gathering around Lean Software Development and I don’t think it’s a coincidence that at the same time many people seem to be falling out of love with Scrum.
At last year’s XPDay there were no sessions on Lean methodologies, this year there are three (the one I’m doing with Matt Wynne, Karl Scotland’s on Kanban, Flow and Cadence and a keynote from the Lean Enterprise Academy).
There’s something about this that makes me feel slightly uneasy. What’s so wrong with Scrum? Well, I’d be the first to stand up and say that Scrum has it’s failings – they’re the same reasons it’s become so popular. The simplicity, clear definition and business-friendliness of Scrum make it easy to sell (arguably, unlike XP) and waterfall or traditional development dynamics (project manager, use-case, Gantt chart, meeting) can be easily translated into Scrum’s characteristics (scrum master, story, burn down, stand up). Unfortunately all too often this is what happens resulting in what’s become know as Cargo Cult Agile, or WAgile, the underlying concepts and belief systems being mostly ignored. When I did Mike Cohn’s Certified Scum Master Course his principle lesson was “Inspect and Adapt” which he repeated endlessly throughout the course, but when I look on the Wikipedia entry for Scrum this doesn’t even get a mention so it’s easy to see how this happens.
So what’s so different about Lean? Well for a start there aren’t any clearly defined rules, but instead principles such as Eliminate Waste and Build Integrity In. However it strikes me that there’s just as much room for abuse with concepts such as Kanban, Minimally Marketable Features, Cumulative Flow Diagrams and so on. Consider this exert from The Toyota Way (which I’ve shamelessly stolen from James Shore’s article on Kanban Systems – I am actually reading this book as we speak, but haven’t got very far yet):
“…TPS experts get very impatient and even irritated when they hear people rave and focus on kanban as if it is the Toyota Production System. Kanban is a fascinating concept and it is fun to watch… When is the kanban triggered? How are the quantities calculated? What do you do if a kanban gets lost? But that is not the point… The challenge is to develop a learning organization that will find ways to reduce the number of and thereby reduce and finally eliminate the inventory buffer… So kanban is something you strive to get rid of, not to be proud of.”
Many of the reasons people aren’t being as succesful as they’d like with Scrum are exactly the same reasons they won’t be any more successful with any other methodology. People tend to focus on tools because it’s a lot easier than trying to tackle the often very difficult, challenging and more fundamental problems they grew from. Real change is hard and takes time, a very long time in some cases.
If you’re failing with Scrum don’t think lean, kanban, extreme programming or any other colour of agile will save you. Essentially if you’re failing it’s because you’re doing it wrong. However if you’ve found Scrum is working really well for you and has brought enormous benefits maybe you should come to the talk Matt and I are doing at XPDay and see how we evolved to a more lean process.
Update: since writing this James Shore has written a frighteningly similar post on The Decline and Fall of Agile, but I guess great minds think alike huh? ;-P
