You are on page 1of 11

November 2004

An Introduction to Agile Software Development
by Victor Szalvay, co-founder
Danube Technologies, Inc.
Web site: http://www.danube.com
Web log: http://danube.com/blog/
12011 Bel-Red Rd. Suite 201
Bellevue, WA 98005
(425) 688-0888, ext. 812

Introduction
This paper is an introduction to the Agile school of software development, and is primarily
targeted at IT managers and CXOs with an interest in improving development productivity.
What is Agile? How can Agile help improve my organization? First, I introduce the two broad
schools of thought when it comes to software development: traditional sequential, a.k.a. “the
waterfall method”, and iterative methods of which Agile is a subset. My objective is to
demonstrate the short-comings of the waterfall approach while providing a solution in iterative,
and more specifically, Agile methods.

Part I – Shortcomings of Traditional Waterfall Approach
The essence of waterfall software development is that complex software systems can be built in a
sequential, phase-wise manner where all of the requirements are gathered at the beginning, all of
the design is completed next, and finally the master design is implemented into production
quality software. This approach holds that complex systems can be built in a single pass, without
going back and revisiting requirements or design ideas in light of changing business or
technology conditions. It was first introduced in an article written by Winston Royce in 1970,
primarily intended for use in government projects1.
Waterfall equates software development to a production line conveyor belt.
“Requirements analysts” compile the system specifications until they pass the finished
requirements specification document to “software designers” who plan the software system and
create diagrams documenting how the code should be written. The design diagrams are then
passed to the “developers” who implement the code from the design (See Figure 1).
Under the waterfall approach, traditional IT managers have made valiant efforts to craft
and adhere to large-scale development plans. These plans are typically laid out in advance of
development projects using Gantt or PERT charts to map detailed tasks and dependencies for
each member of the development group months or years down the line. However, studies of past
software projects show that only 9% to 16% are considered on-time and on-budget2. In this
article, I attempt to summarize current thinking among computer scientists on why waterfall fails
in so many cases. I also explore a leading alternative to waterfall: “Agile” methods that focus on

Copyright © 2004 Danube Technologies, Inc. All rights reserved.

Code 4. the SRS is sent over the fence to the designers while the requirements analysts go to work on the next project. Page 2 incremental and iterative development where requirements. user interface standards. Once finished. and in many waterfall projects this can be very costly. the first block represents the requirements analysis phase of a software development project. What if your business needs are still emerging and certain aspects of the system are rapidly changing or cannot be defined yet? Business climates and objectives often change Copyright © 2004 Danube Technologies.999% of the time. Figure 1 Traditional Methods: sequential phased approach Project 1. Deploy Analysis Design Phase Up-front Requirements Analysis What are requirements? From the stakeholder’s perspective. Requirements define what developers are to build. Requirements 2. scalability and concurrent user support. One of the biggest problems with waterfall is that it assumes that all project requirements can be accurately gathered at the beginning of the project. Inc. the system must have a web site with e-commerce capability that can handle 10. implementation. By virtue of a requirements change. it is inevitable that attempts at up-front requirements specification will leave out some very important details simply because the stakeholders cannot tell developers everything about the system at the beginning of the project. Test 5. In Figure 1. In fact. or the system must be accessible 99. The cost of change in a waterfall project increases exponentially over time because the developer is forced to make any and all project decisions at the beginning of the project. Architecture & 3. Analysts slave for weeks or months compiling everything they can gleam about the proposed system into comprehensive “Software Requirements Specification” (SRS) documents. Imagine a scenario where you engage a software group to build a critical software system. For example. . which will in turn affect any implementation and test strategies. browser or OS support.3 This means that the requirements typically change outside of the requirements phase in the form of “change orders”. and testing continue throughout the project lifecycle. All rights reserved.000 purchases per hour. design. the requirements are the features and specifications of the system. Do you think you could provide every last detail the developers need to know right off the bat? I have yet to encounter such a customer and I am hard pressed to think I ever will. user roles and restrictions. consider the areas that must be addressed: business rules and exceptions. the intricately planned design can be affected dramatically. As a start.

when I draw a picture I need to see the drawing as I progress. functional software. the requirements are solicited from the user and some time later the finished product is presented to the user. Page 3 rapidly. I often use a “drawing” analogy to help explain this effect. and artistry. Also consider that the building blocks of software projects is usually other software systems (e. however. Although I’m a terrible artist. and those systems that act as building blocks contain bugs and cannot be relied on with certainty. This is entirely unnatural because customers find it difficult to specify software perfectly without seeing it evolve and progress. People need to see and feel something before they really know what they want. changing. Bridge building relies on physical and mathematical laws. All rights reserved.g. Because the foundations of software development are inherently unstable and unreliable.. waterfall projects allocate copious effort detailing every possible risk. . Markets are forcing the software development community to respond with flexible development plans that flatten the cost of change. If I tried to close my eyes and draw the same picture. Waterfall is an “over the fence” approach. Software Development is more like New Product Development than Manufacturing Software development is a highly complex field with countless variables impacting the system. Since the test itself had not been released yet. Software development is innovation. it would prove far less successful. In fact. database platforms. As a result. and emerging requirements presents a very large challenge to waterfall projects because by definition all requirements must be captured up-front at the risk of costly changes later. Software development. The waterfall methodology assumes that up-front planning is enough to take into account all variables that could impact the development process. each foray into a development project presents new and difficult challenges that cannot be overcome with one-size-fits-all. Can you afford to lock your business into a rigid long-term project where the cost of change grows exponentially? For example. But the system had to be done shortly after the tests were released. organizations developing software must realize variables exist that are largely outside of management control. and contingency. It is therefore fair to say that software development is more akin to new product research and development than it is to assembly-line style manufacturing. programming languages. mitigation plan. The “I’ll Know it When I See It” (IKIWISI) law says that software development customers can better describe what they really want after seeing and trying working. a national test preparation organization commissioned my company to build a simulator for an upcoming standardized test. has no laws or clear certainties on which to build. cookie- cutter solutions4. But this is what waterfall asks customers to do: specify the entire system without having a chance to periodically see the progress and make adjustments to the requirements as needed. discovery. Inc. especially in today’s age of instant information. But is it possible Copyright © 2004 Danube Technologies. the types of questions that would appear on the test were unknown when we started development. software is almost always flawed or sub-optimized. etc.). All software systems are imperfect because they cannot be built with mathematical or physical certainty. The problem of undefined.

. or iterations. The first step is X. over a project lifecycle to properly address complexities and risk factors. On the contrary. defined processes can be established that.5 Waterfall therefore equates software development to an assembly line. For example. the benefits of which other new product development industries have been reaping for decades. Winston Royce (of waterfall process fame) later noted that his ideas were incorrectly interpreted and that a “single pass” framework would never work (his article actually advocates at least a second pass). The idea of revisiting phases over and over is called “incremental and iterative development” (IID).Iterative and Agile methods Incremental and Iterative Development The simple ability to revisit the “phases” of development dramatically improves project efficiency. medical researchers could simply plug variables into equations to discover new medicines. Almost no software system is so simple that the development can be entirely scripted from beginning to end.7 But longstanding insistence from IT managers to categorize software development as a straightforward assembly line progression has kept the software industry from evolving to better methods. Inc. This concept of iterative development hails from the “lean development” era of the 1980s described above where Japanese auto makers made tremendous efficiency and innovation increases simply by removing the phased. and the result is always Z. IID processes continually capture the requirements iteration after iteration Interestingly. Can research really be relegated to a series of steps that when performed in sequence result in a new product? If this formulaic approach were adequate. where prototypes were developed for short-term milestones (see Figure 2). sequential approach and implementing an iterative approach. Fujitsu. in the 1990s sequential.8 IID allows for multiple “passes”. since the late 1970s product development companies lead by Toyota. Part II . Each Copyright © 2004 Danube Technologies. Canon. with IID requirements is an ongoing process that is periodically revisited. As new requirements surface and as the scope changes. It’s ironic that a cutting edge technology field like software is so far behind more traditional engineering fields in terms of development methods. The development lifecycle is cut up into increments or “iterations” and each iteration touches on each of the traditional “phases” of development. and NEC. The inherent uncertainty and complexity in all software projects requires an adaptive development plan to cope with uncertainty and a high number of unknown variables. 6 The results were a dramatic improvement in cost and development time to market and ultimately lead to the popular rise of “lean development” and “just-in-time manufacturing”. holistic approach where the traditional phases of development overlap throughout the product lifecycle. supplanted the sequential “Phased Program Planning” (PPP) approach to new product development with a flexible. the second is Y. Honda. waterfall-style approaches to new product development were effectively abandoned outside the software development industry. Page 4 to predict any and all variables that could possibly affect a software project? The empirical answer is “no” considering the limited success of waterfall projects. HP. All rights reserved. when used sequentially. Following the lead of Japanese auto makers. result in a successful project each time. 3M.

a concise summary of Agile values. pp.. Agile methods: Embracing Change Agile methods stress productivity and values over heavy-weight process overhead and artifacts. and they further increase the iterative nature of the software lifecycle by tightening design-code-test loop to at least once a day (if not much more frequently) as opposed to once per iteration. All rights reserved. To accomplish this “flatter” cost of change curve. The Agile Manifesto10. Inc. This “concurrent” development approach created an atmosphere of trial-and-error experimentation and learning that ultimately broke down the status quo and led to efficient innovation. Spiral. . Takeuchi and I. and implementation cycle was revisited for each short-term milestone. the requirements.9 Although direct analogies between industries are never seamless. 1986. Agile methods promote an iterative mechanism for producing software. small increments of functional code are produced according to immediate business need. was written and signed in 2001 although Agile methods have existed since the early 90s. the success of lean development has influenced a broad class of “iterative” software methods including the Unified Process. Harvard Business Rev. Author and speaker Martin Fowler describes testing and continuous integration as the “enabling” Agile practices that allow for the advantages gained. Beck’s model espouses that the cost of change can be inexpensive even late in the project lifecycle while maintaining or increasing system quality12. “Test driven development” is a quality-first approach where developer tests (called unit tests) are written prior to the functional code itself. Jan. Page 5 phase was actually a layer that continued throughout the entire development lifecycle. Nonaka. Beck’s idealistic “flat” cost of change curve has since been revised and softened by Alister Cockburn 13 and Scott Ambler 14 to reflect modern corporate realities. like rapid production and minimum up-front design 15. and Agile methods. Agile methods promote a number of engineering practices that enable cost effective change. Evo. Nevertheless. 137-146. It is the role of the automated test suite built around the rapidly evolving code to act as a harness that Copyright © 2004 Danube Technologies. Agile visionary Kent Beck challenged the traditional cost of change curve evidenced by Barry Boehm11 over twenty years ago. Figure 2 Iterative approach: Overlapping phases of development Phase 1 2 3 4 5 6 Source: Adapted from H. Rather than focusing a lot of effort on big up front design analysis. "The New New Product Development Game". design. Agile ideals can be applied to reduce the cost of change throughout the software lifecycle even if the cost of change is not perfectly flat.

for example. In Copyright © 2004 Danube Technologies. Agile project management approaches balance the four variables in software development while keeping in mind the limits associated with new product development. are too difficult to script and define. Page 6 allows developers to make aggressive code changes without fear of undetected regression failure. the manufacturing industry has long known that certain chemical processes. in the Scrum process. Agile Project Management: Empirical Process Scrum. a popular Agile project management method. Risk factors and emerging requirements complicate software development to a point where defined processes fall short. Figure 3: Cost of Change Curves Cost of Change Boehm Cockburn/Ambler Beck Project Lifecycle Progress Object technology and modern integrated development environments (IDEs) boasting built-in testing and refactoring mechanisms negate the expensive Boehm cost of change curve and allow for the cheap change. Inc. In fact. introduced the concept of empirical process control for the management of complex. All rights reserved. an empirical or adaptive management approach is employed to measure and adjust the chemical process periodically to achieve the desired outcome.16 As a result. project plans are continuously inspected and adapted based on the empirical reality of the project. Although it has been attempted in the past. even late in the project lifecycle. Instead. . changing software projects. Scrum holds that straightforward defined processes alone cannot be used to effectively manage complex and dynamic software projects. there cannot be a single exhaustive library of defined processes to handle every situation that could possibly surface during a software project.

. However software development cannot be described by a simple linear process because it cannot be predicted accurately in advance. It is therefore unreasonable to assume that management can control all four of these factors. Available money impacts the amount of effort put into the system. Page 7 software development there are four broad control factors. and resources are utilized efficiently to avoid “queuing” (see “queuing theory” and the theory of constraints). Agile values a high visibility and customer involvement. Second. Consider for example the “small batch” principle: things produced in smaller batches are of higher quality and efficiency because the feedback loop is short. middle and upper management often assumes that all four of these factors could be dictated to the development team under the waterfall approach. § Requirements – The scope of the work that needs to be done can be increased or decreased to affect the project. Inc. The frequent demonstration and release of software common in Agile approaches Copyright © 2004 Danube Technologies. Agile methods have their conceptual roots in the Japanese manufacturing productivity boom of the 1980s 19. organizations can reduce risks by leaving options open to decide at a better time when more accurate information is available. when one changes at least one other factor must also change. In reality. management can pick values for three of the four factors at most. Many software development organizations that implement Agile software development are finding they get something they never expected: options. § Cost – or Effort.20 Agile Requirements: A Focus on Business ROI Agile projects avoid “up-front” requirements gathering for the reasons stated above: customers cannot effectively produce all requirements in high enough detail for implementation to occur at the beginning of a project. § Quality – Cut corners by reducing quality. thereby reducing the risks associated with large integrations at the tail end of projects. the concept of frequent or continuous integration keep software development teams synchronized. Agile methods encourage delaying irreversible decisions until the last responsible moment. Customers may not want to make decisions about the system until they have more information. and the development process dictates the fourth. Lean Thinking Another effective way to analyze how Agile methods increase efficiencies is to apply lean manufacturing principles to software development. Teams can work independently for a while but the code base never diverges for long periods of time. Rather than locking into decisions at the beginning of a project. § Schedule – A software project is impacted as the timeline is changed. These factors are interconnected. Although cross-industry analysis can be tenuous.17 Because software development is often considered a sequential. Third. linear process. All rights reserved.18 The highly complex and uncertain nature of software development makes this expectation of full control unrealistic. controls can be adjusted more frequently.

The trendline can then be Copyright © 2004 Danube Technologies. Quite the contrary. Page 8 gives customers a chance to “try software” periodically and provide feedback. the development team must bring technical risks to the customer. Agile processes like Scrum and XP use a concept called velocity. Decisions can be delayed to some future iteration when better information or technology is available to optimize the choice. Agile helps companies produce the “right product”. But we already know that we cannot plan for everything in advance. Inc. and (luckily) a few weeks before the product launch a new version was released by one of the database vendors that solved our problem. Because we can change direction rapidly (every iteration) and the cost of change is low. This is especially the case with small businesses where domain experts wear many hats and often cannot commit to two or three months of straight requirements analysis. a trendline can be established through the points to create a velocity (work the team can complete per iteration). and the Y-axis represents the total estimated effort remaining for the backlog. Convergence with Agile One of the most commonly asked questions by those examining Agile is. The Agile answer is to examine project progress empirically. Once a team has established a velocity. How do Agile projects prioritize work? A study by the Standish Group shows that in typical software systems 45 percent of the features are never actually used by users and another 19% are only rare used. Many organizations are not fully staffed with business analysts cranking out reams of requirements specs. there is a valuable opportunity for the customer to re-examine business factors at the beginning of each iteration to select features for inclusion in the current iteration according to business ROI. we recently delayed selecting a database package for an application because some of the desired features were not available at that time in the options we had to choose from. “how do you know when the software will be finished if there’s no up-front plan?” and the obligatory follow up question. a Project Burndown Chart can be utilized to estimate the eventual conclusion of an estimated backlog of work. All rights reserved. One of the biggest advantages to IID is that work can begin before all of the requirements are known.21 This is largely because the unused features were specified in some up- front plan before the ratio of their cost to value was considered or even knowable. We therefore built the system in a database independent manner. Each point of the chart in Figure 4 represents an iteration (or Sprint in Scrum). Focusing on high business value features first is therefore a critical component of efficient Agile development. rather than trying to guess how things might shape up a priori. IID is ideally suited then to take on bite-sized chunks of requirements that the customer can easily digest. in our experience often the bottleneck in the development process has been the lack of availability of customer domain experts for detailed requirements analysis. “how can we budget for such a project?” It sounds a bit scary: let’s start working in short iterative cycles that yield demonstrable software without actually planning everything in advance. An iterative approach allows customers to delay decisions as well. As iterations progress. . which is the amount of estimated effort a team can complete in a time-boxed iteration. but in the end it is the customer that decides what the development team builds. For example. Of course.

if the timeline and cost variables are fixed. All rights reserved. schedule. For complex problems like project convergence. the features built should always be high quality. In this case. . scope. then the scope of the work must be variable or the definition of the scope must be at a high level so the robustness of each feature can be negotiated. quality). the product scope had been adjusted down so that the project could be completed by the budgeted 20th iteration. To answer the questions above. Figure 4: Product Burndown Chart with Velocity 1200 Velocity: 5 story pts per sprint 1000 800 Story Points 600 Velocity: 25 story pts per sprint 400 200 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Sprints Figure 3 is a Product Burndown Chart representation of a typical project. especially considering the Standish report cited above that nearly 65% of features are never or rarely used in reality. Going to production with high priority features is better than never going to production at all. Copyright © 2004 Danube Technologies. work will proceed on the highest priority requirements first to ensure that the most important things get done before a deadline or the money runs out. There is no guarantee all features will be built. And quality is non-negotiable. Through the first nine iterations. the project’s “burndown” trendline indicated an X-intercept well into the future (off the charts!). Inc. Page 9 extrapolated to determine the X-intercept.22 That is. Agile methods tell us that the customer cannot specify all four of the software development variables (cost. adhering to strict code and testing standards. but it is certain that the highest priority features will go into production and that they will be built well. By iteration ten. which represents the empirical estimate of the completion date. perhaps the reduction in scope was accompanied by the removal of critical impediments to efficient progress. Notice also that the velocity (slope) changed for the better.

Addison-Wesley. Addison-Wesley. sometimes I wish I could ask the author for a list of good books or articles to further my knowledge on the subject. 2001. much of this article leans on the Poppendieck’s work. Craig Larman. 2003. Kent Beck. It contains the most compelling case for agile over sequential development I have yet to uncover. This book should be read by any manager interested in Agile. tried and true history of waterfall software development is incorrect. and the Unified Process. Addison-Wesley. Prentice Hall. It also nicely summarizes and contrasts some of the major Agile/Iterative approaches such as Scrum. Inc. “Extreme Programming Explained: Embrace Change”. All rights reserved. a highly effective agile project management method. It also delivers outstanding tools for implementing agile. “Agile & Iterative Development: A Manager’s Guide”. Theory and rhetoric are nice. XP. 2000. Agile is simply the latest theory that is widely replacing the waterfall approach that itself will change and evolve well into the future. So what do your developers do differently in agile? Extreme Programming (or XP) advocates engineering principles such as pair programming and test driven development and Beck’s book is the de facto authority. “Lean Software Development An Agile Toolkit”. Page 10 Conclusion But does Agile/IID work? Of course the proof is always in the pudding. 23 Standish Chairman Jim Johnson attributes the improvement to smaller projects using iterative processes as opposed to the waterfall method.25 References After reading an introductory article. software development is not built on thousands of years of trial and error. Standish reported a 31% failure rate that has improved to 15% in 2004. “Agile Software Development with Scrum”. Unlike bridge building. This book is about “Scrum”. In 1994. Tom Poppendieck. It has the most comprehensive empirical evidence for Agile/Iterative of any book currently on the market. Although waterfall is often referred to as “traditional”. Mike Beedle. but how do you do agile? Scrum is a very good place to start on the management side. and the most recent 2004 Standish Group CHAOS report on the success of software projects shows a dramatic improvement in the failure rate of software projects. Ken Schwaber. This book is outstanding and each page seems to offer a valuable nugget of information. Mary Poppendieck. and is therefore in a rapidly evolving infancy as an engineering discipline. 2003. software engineering has had a very short history relative to other engineering disciplines.24 The notion that Agile is a radical deviation from the long established. . Below are a few good starting points for anyone interested in learning more about agile software development. Copyright © 2004 Danube Technologies.

20 Mary Poppendieck. 8 Craig Larman. 2001. see: Mary Poppendieck. Addison-Wesley. 1970.php 3 Kent Beck. 2003. Developers write only robust tests and supercomputers will write and compile code until all tests pass. http://martinfowler. 2004. Inc. Nonaka. Software developers should always aim for high quality software. and for the same ethical implications software developers resent this charge. Harvard Business Review. 2004. IEEE CS Press. by Scott Ambler. pp. it is debatable whether Quality is really an adjustable factor. “Lean Software Development An Agile Toolkit”. Proc.com/articleshttps://www.scribd.com/designDead. Prentice Hall. Software Magazine and Weisner Publishing.. 2004. As professionals. “Agile Software Development with Scrum”. Prentice Hall. Tom Poppendieck. p 28. by Scott Ambler. Tom Poppendieck. 2002. http://www. Addison-Wesley. June 2004.. “Lean Software Development An Agile Toolkit”.standishgroup. IEEE CS Press. 3rd ed. clustered supercomputing. 21 Jim Johnson. Prentice Hall PTR. Inc. http://www1. Takeuchi. Agile Modeling Essays excerpted from the book “The Object Primer. Surgeons or lawyers would be sued for malpractice.com/L. Computer. 7 For more on how lean development influences agile software development. January 1986. by Alistair Cockburn.html . software developers find it very objectionable when asked to skimp on quality. H. 14 Examining the Cost of Change Curve. 2001.agilemanifesto. Addison-Wesley. http://www1. “Iterative and Incremental Development: A Brief History”. 137-146.. . 89-94 5 Standish Group International. Copyright © 2004 Danube Technologies. Addison-Wesley. “Is Design Dead”. 18 Kent Beck. 100-101 17 In fact.cfm?Doc=newsletter/2004-01-15/Standish. “Chaos Chronicals”. pp. “Extreme Programming Explained: Embrace Change”. 10 The Agile Manifesto is online at http://www.com//sample_research/chaos_1994_1. 15-19. pp. January 1986. 2003. Inc. 2003. Mike Beedle. Victor R. Published Keynote Third International Conference on Extreme Programming. 2000. Westcon. September 2000. All rights reserved. 13 Reexamining the Cost of Change Curve. XP Magazine. 1994. Nonaka. Takeuchi. 2000. “Chaos Chronicles”. p.php 6 I. pp. Mike Beedle. “Agile Software Development with Scrum”. 16 Ken Schwaber. 18-19. Cambridge University Press. Inc. 48.standishgroup. p 32. pp. 1994. 22 Mary Poppendieck. 1981. pp. “The New New Product Development Game”. 24 “Standish: Project Success Rates Improved Over 10 Years”. 12 Kent Beck.softwaremag. It’s Your Job!”. H.org/ 11 Barry Boehm. 2000. 9 I. 15 Martin Fowler. pp. 23 Standish Group International. Basili. “ROI. 4 Ken Schwaber. Harvard Business Review. period. year 2000. “Chaos Chronicals”. Tom Poppendieck. Addison-Wesley. “The New New Product Development Game”. 328-339. 2 Standish Group International. Page 11 1 Winston Royce. “Software Engineering Economics”. 19 The Scrum Agile method has its roots in the Nonaka-Takeuchi article referenced above. “Lean Software Development An Agile Toolkit”. Addison-Wesley. “Extreme Programming Explained: Embrace Change”.com//sample_research/chaos_1994_1. 137-146. “Managing the Development of Large Software Systems”.: Agile Model-Driven Development with UML2”. “Extreme Programming Explained: Embrace Change”. 25 Software architect Michael James has a semi-serious theory that “test-only” development will soon be feasible with the advances in cheap.