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.

In fact. Once finished. the SRS is sent over the fence to the designers while the requirements analysts go to work on the next project. 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. and in many waterfall projects this can be very costly. Requirements 2.000 purchases per hour. or the system must be accessible 99.3 This means that the requirements typically change outside of the requirements phase in the form of “change orders”. Inc. For example. 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. By virtue of a requirements change. the system must have a web site with e-commerce capability that can handle 10. user roles and restrictions. As a start. and testing continue throughout the project lifecycle. Figure 1 Traditional Methods: sequential phased approach Project 1. Analysts slave for weeks or months compiling everything they can gleam about the proposed system into comprehensive “Software Requirements Specification” (SRS) documents. Deploy Analysis Design Phase Up-front Requirements Analysis What are requirements? From the stakeholder’s perspective. the first block represents the requirements analysis phase of a software development project. the requirements are the features and specifications of the system. Page 2 incremental and iterative development where requirements. user interface standards. 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. Imagine a scenario where you engage a software group to build a critical software system. All rights reserved. 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. . which will in turn affect any implementation and test strategies. In Figure 1.999% of the time. design. consider the areas that must be addressed: business rules and exceptions. implementation. the intricately planned design can be affected dramatically. Architecture & 3. browser or OS support. Test 5. scalability and concurrent user support. Requirements define what developers are to build. 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. Code 4.

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

Canon.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. and the result is always Z. over a project lifecycle to properly address complexities and risk factors. Almost no software system is so simple that the development can be entirely scripted from beginning to end.8 IID allows for multiple “passes”. medical researchers could simply plug variables into equations to discover new medicines. The development lifecycle is cut up into increments or “iterations” and each iteration touches on each of the traditional “phases” of development.Iterative and Agile methods Incremental and Iterative Development The simple ability to revisit the “phases” of development dramatically improves project efficiency. holistic approach where the traditional phases of development overlap throughout the product lifecycle. since the late 1970s product development companies lead by Toyota. when used sequentially. Honda. result in a successful project each time. sequential approach and implementing an iterative approach. Fujitsu. supplanted the sequential “Phased Program Planning” (PPP) approach to new product development with a flexible. 3M. For example. HP. Each Copyright © 2004 Danube Technologies. and NEC. As new requirements surface and as the scope changes. with IID requirements is an ongoing process that is periodically revisited. waterfall-style approaches to new product development were effectively abandoned outside the software development industry. The idea of revisiting phases over and over is called “incremental and iterative development” (IID). or iterations. the benefits of which other new product development industries have been reaping for decades. IID processes continually capture the requirements iteration after iteration Interestingly. where prototypes were developed for short-term milestones (see Figure 2). The first step is X. Part II . 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). 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”. On the contrary. 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. the second is Y. Following the lead of Japanese auto makers. defined processes can be established that.5 Waterfall therefore equates software development to an assembly line. in the 1990s sequential. 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. 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. All rights reserved. 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. . Inc. It’s ironic that a cutting edge technology field like software is so far behind more traditional engineering fields in terms of development methods.

Takeuchi and I. like rapid production and minimum up-front design 15. “Test driven development” is a quality-first approach where developer tests (called unit tests) are written prior to the functional code itself. 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. the success of lean development has influenced a broad class of “iterative” software methods including the Unified Process. Agile methods: Embracing Change Agile methods stress productivity and values over heavy-weight process overhead and artifacts. Spiral. and implementation cycle was revisited for each short-term milestone. Jan. 1986.9 Although direct analogies between industries are never seamless. Figure 2 Iterative approach: Overlapping phases of development Phase 1 2 3 4 5 6 Source: Adapted from H. Nevertheless. 137-146. All rights reserved. "The New New Product Development Game". a concise summary of Agile values. Rather than focusing a lot of effort on big up front design analysis. 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. Beck’s model espouses that the cost of change can be inexpensive even late in the project lifecycle while maintaining or increasing system quality12. Page 5 phase was actually a layer that continued throughout the entire development lifecycle.. Author and speaker Martin Fowler describes testing and continuous integration as the “enabling” Agile practices that allow for the advantages gained. 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. To accomplish this “flatter” cost of change curve. Agile methods promote a number of engineering practices that enable cost effective change. Agile methods promote an iterative mechanism for producing software. Agile visionary Kent Beck challenged the traditional cost of change curve evidenced by Barry Boehm11 over twenty years ago. was written and signed in 2001 although Agile methods have existed since the early 90s. The Agile Manifesto10. Inc. pp. 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. Evo. Harvard Business Rev. small increments of functional code are produced according to immediate business need. and Agile methods. . 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. the requirements.

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

Available money impacts the amount of effort put into the system.17 Because software development is often considered a sequential. Agile methods have their conceptual roots in the Japanese manufacturing productivity boom of the 1980s 19.20 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. controls can be adjusted more frequently. Rather than locking into decisions at the beginning of a project. Agile methods encourage delaying irreversible decisions until the last responsible moment. Agile values a high visibility and customer involvement. Teams can work independently for a while but the code base never diverges for long periods of time. Second. These factors are interconnected. Customers may not want to make decisions about the system until they have more information. § Requirements – The scope of the work that needs to be done can be increased or decreased to affect the project. . organizations can reduce risks by leaving options open to decide at a better time when more accurate information is available. middle and upper management often assumes that all four of these factors could be dictated to the development team under the waterfall approach. thereby reducing the risks associated with large integrations at the tail end of projects. Many software development organizations that implement Agile software development are finding they get something they never expected: options. Lean Thinking Another effective way to analyze how Agile methods increase efficiencies is to apply lean manufacturing principles to software development. § Quality – Cut corners by reducing quality. when one changes at least one other factor must also change. and the development process dictates the fourth. Third. Inc. and resources are utilized efficiently to avoid “queuing” (see “queuing theory” and the theory of constraints). Page 7 software development there are four broad control factors. Consider for example the “small batch” principle: things produced in smaller batches are of higher quality and efficiency because the feedback loop is short. Although cross-industry analysis can be tenuous. § Cost – or Effort. All rights reserved. However software development cannot be described by a simple linear process because it cannot be predicted accurately in advance. In reality. The frequent demonstration and release of software common in Agile approaches Copyright © 2004 Danube Technologies. linear process. the concept of frequent or continuous integration keep software development teams synchronized.18 The highly complex and uncertain nature of software development makes this expectation of full control unrealistic. management can pick values for three of the four factors at most. § Schedule – A software project is impacted as the timeline is changed. It is therefore unreasonable to assume that management can control all four of these factors.

Quite the contrary. The Agile answer is to examine project progress empirically. in our experience often the bottleneck in the development process has been the lack of availability of customer domain experts for detailed requirements analysis. but in the end it is the customer that decides what the development team builds. Inc. For example. An iterative approach allows customers to delay decisions as well. Many organizations are not fully staffed with business analysts cranking out reams of requirements specs. Focusing on high business value features first is therefore a critical component of efficient Agile development. Each point of the chart in Figure 4 represents an iteration (or Sprint in Scrum). “how do you know when the software will be finished if there’s no up-front plan?” and the obligatory follow up question. We therefore built the system in a database independent manner. IID is ideally suited then to take on bite-sized chunks of requirements that the customer can easily digest. All rights reserved. But we already know that we cannot plan for everything in advance. Page 8 gives customers a chance to “try software” periodically and provide feedback. Agile helps companies produce the “right product”. 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. Once a team has established a velocity. 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 Project Burndown Chart can be utilized to estimate the eventual conclusion of an estimated backlog of work. rather than trying to guess how things might shape up a priori. the development team must bring technical risks to the customer. which is the amount of estimated effort a team can complete in a time-boxed iteration. and the Y-axis represents the total estimated effort remaining for the backlog.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. One of the biggest advantages to IID is that work can begin before all of the requirements are known. a trendline can be established through the points to create a velocity (work the team can complete per iteration). Agile processes like Scrum and XP use a concept called velocity. The trendline can then be Copyright © 2004 Danube Technologies. 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. 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. As iterations progress. Of course. Because we can change direction rapidly (every iteration) and the cost of change is low. and (luckily) a few weeks before the product launch a new version was released by one of the database vendors that solved our problem. Convergence with Agile One of the most commonly asked questions by those examining Agile is. “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. Decisions can be delayed to some future iteration when better information or technology is available to optimize the choice. .

By iteration ten. the product scope had been adjusted down so that the project could be completed by the budgeted 20th iteration. the project’s “burndown” trendline indicated an X-intercept well into the future (off the charts!). Page 9 extrapolated to determine the X-intercept. scope. To answer the questions above. Going to production with high priority features is better than never going to production at all. And quality is non-negotiable. 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. perhaps the reduction in scope was accompanied by the removal of critical impediments to efficient progress. Copyright © 2004 Danube Technologies.22 That is. Notice also that the velocity (slope) changed for the better. adhering to strict code and testing standards. quality). if the timeline and cost variables are fixed. schedule. which represents the empirical estimate of the completion date. 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. 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. Through the first nine iterations. There is no guarantee all features will be built. Agile methods tell us that the customer cannot specify all four of the software development variables (cost. but it is certain that the highest priority features will go into production and that they will be built well. All rights reserved. In this case. the features built should always be high quality. . Inc. especially considering the Standish report cited above that nearly 65% of features are never or rarely used in reality. For complex problems like project convergence.

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

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