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.

999% of the time. or the system must be accessible 99. Once finished. implementation. . the SRS is sent over the fence to the designers while the requirements analysts go to work on the next project. design. Deploy Analysis Design Phase Up-front Requirements Analysis What are requirements? From the stakeholder’s perspective. All rights reserved. browser or OS support. the intricately planned design can be affected dramatically. Test 5. the requirements are the features and specifications of the system. Page 2 incremental and iterative development where requirements. scalability and concurrent user support.000 purchases per hour. In fact. 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. Figure 1 Traditional Methods: sequential phased approach Project 1. Imagine a scenario where you engage a software group to build a critical software system. consider the areas that must be addressed: business rules and exceptions. In Figure 1. Analysts slave for weeks or months compiling everything they can gleam about the proposed system into comprehensive “Software Requirements Specification” (SRS) documents. 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. 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. Requirements 2. which will in turn affect any implementation and test strategies. As a start. the first block represents the requirements analysis phase of a software development project. and in many waterfall projects this can be very costly. By virtue of a requirements change. user roles and restrictions.3 This means that the requirements typically change outside of the requirements phase in the form of “change orders”. Requirements define what developers are to build. Architecture & 3. Inc. and testing continue throughout the project lifecycle. 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. 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. the system must have a web site with e-commerce capability that can handle 10. Code 4. user interface standards. For example.

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

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

the requirements. 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. 137-146. was written and signed in 2001 although Agile methods have existed since the early 90s. All rights reserved. Spiral. design. Agile methods: Embracing Change Agile methods stress productivity and values over heavy-weight process overhead and artifacts. a concise summary of Agile values.. Agile methods promote an iterative mechanism for producing software. Harvard Business Rev. Author and speaker Martin Fowler describes testing and continuous integration as the “enabling” Agile practices that allow for the advantages gained. Agile methods promote a number of engineering practices that enable cost effective change. The Agile Manifesto10. like rapid production and minimum up-front design 15. Takeuchi and I.9 Although direct analogies between industries are never seamless. pp. small increments of functional code are produced according to immediate business need. Nonaka. 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. 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. 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. "The New New Product Development Game". 1986. the success of lean development has influenced a broad class of “iterative” software methods including the Unified Process. . 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. “Test driven development” is a quality-first approach where developer tests (called unit tests) are written prior to the functional code itself. Beck’s model espouses that the cost of change can be inexpensive even late in the project lifecycle while maintaining or increasing system quality12. Inc. To accomplish this “flatter” cost of change curve. Jan. 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. Page 5 phase was actually a layer that continued throughout the entire development lifecycle. Rather than focusing a lot of effort on big up front design analysis. and Agile methods. Nevertheless. and implementation cycle was revisited for each short-term milestone. Evo.

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

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

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

Notice also that the velocity (slope) changed for the better. All rights reserved. By iteration ten. the features built should always be high quality. 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. if the timeline and cost variables are fixed. 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. In this case. And quality is non-negotiable. To answer the questions above. 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. Going to production with high priority features is better than never going to production at all. Inc. Page 9 extrapolated to determine the X-intercept. but it is certain that the highest priority features will go into production and that they will be built well. the project’s “burndown” trendline indicated an X-intercept well into the future (off the charts!). scope. 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. schedule.22 That is. quality). adhering to strict code and testing standards. perhaps the reduction in scope was accompanied by the removal of critical impediments to efficient progress. especially considering the Standish report cited above that nearly 65% of features are never or rarely used in reality. which represents the empirical estimate of the completion date. the product scope had been adjusted down so that the project could be completed by the budgeted 20th iteration. Through the first nine iterations. Copyright © 2004 Danube Technologies. For complex problems like project convergence.

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

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