An Introduction to Agile Emergent Architecture – Always Intentional

An Introduction to Agile Emergent Architecture: Always Intentional

Let’s Define Architecture
By definition, architecture is about the major pieces of the system^1, and about satisfying the non functional requirements such as availability, usability, compliance, scalability, security, extensibility , maintainability, and all of the typical “ilities.” It’s important to remember that architecture does not cover anything and everything that can be put into one of the above categories — because they would encompass every last bit of the *entire* system. The architecture should not encompass every last corner of the system. Instead, architecture focuses on the very highest level concerns in each of those categories, and nothing more than that. Further, architecture focuses only on the most basic of building blocks in those categories, and nothing more than that. This leaves as much flexibility as possible for later agility and implementation. So, be sure your architecture is focusing only on the highest of concerns. (The lower concerns might be considered such things as design, code, or implementation)

Big Up Front(BUF) Thinking
Legacy software development approaches put a large emphasis on understanding all the requirements for a system up front. This is also known as BRUF (Big Requirements Up Front). The disproven theory behind this thought is that if one understood all the requirements really well up front, then the most optimum architectures, designs, and software code could be created. As such, BRUF has siblings known as BAUF (Big Architecture Up Front) and BDUF (Big Design Up Front). All of these Big Up Front (BUF) approach theories have been disproven for one overall reason: it takes time to gain this BUF understanding, and by the time you do think you and your team have that understanding, the requirements have changed substantially, which then affects that architecture and design as well. What we have learned over the decades is that the requirements, technologies, users, technical team members, and the market for the product change quicker than we can gain that understanding, and certainly quicker than we can implement an architecture or design. In addition, customers often have trouble describing exactly what they want — they tend to have a much better understanding of what they want only after they see the working software. Further, with BUF thinking, usually the people that did the up front thinking are way different than the people implementing that thinking. So downstream from requirements, when changes to the architecture and design are truly needed, because of all of the previous history, BUF thinking, and totally different people involved, changes are extraordinarily costly *and* complex to retrofit. As such, Big Up Front thinking is a model that has been disproven, so it’s unacceptable, and indeed now.. passe.

No Up Front(NUF) Thinking
However, the other end of the spectrum, NUF (No Up Front Thinking), is also unacceptable. You can’t create a cohesive architecture, that is financially viable, without some good upfront thinking. Without at least some good upfront thinking, the architecture turns out haphazard, almost accidental in nature. Some people have called this unintentional architecture, and the name probably fits.

Signs you MIght be a Victim of NUF
One possible sign of NUF is a system rewrite. A system rewrite is almost always a sign of failure ^2, and the two biggest reasons for that failure are inattention to the user marketplace and inattention to continuous technical excellence via architecture, design, testing and coding practices. Inattention to continuous technical excellence creates what is known as technical debt, which is the other tell tale sign of NUF. Massive technical debt generally presents itself as a highly unacceptable amount of bugs, new functionality that takes way longer than it should, or systems that get jettisoned or re-written. Massive technical debt is pretty much a guaranteed outcome of No Up Front Thinking. As such, since massive technical debt has so many bad outcomes, No Up Front Thinking is also unacceptable.

Big Up Front thinking can lead to No Up Front Thinking
Now, before we move on, let me also address another relevant point. BUF thinking can actually lead to NUF. If a BUF architecture is not kept up to date, is not shepherded, or is extremely inflexible to change, people will avoid thinking about and making architectural changes. This results in NUF and massive technical debt. So, a single system can be a victim of both BUF and NUF. It would be really great if there was a way to find that right balance between BUF and NUF… right?

The Better Way: Agile Emergent Architecture
Enter “Emergent Architecture”, a term suggested by Agile thought leaders. One can think of this as Little Up Front(LUF) Architecture, combined with continuous attention to technical excellence. You could even think of it as “Continuous Architecture” if you so desired. With Emergent Architecture, you do just enough, just in time, at the “last responsible moment.” It’s also important to note that Emergent Architecture is also 100% intentional architecture. Architecture doesn’t just “magically appear”.

The Benefits of Emergent Architecture
By architecting at the last responsible moment, you are minimizing the requirement churn damage that accompanies BUF. By architecting at the last responsible moment, you are taking advantage of the latest and greatest technology knowledge. Finally, by architecting at the last responsible moment, you are very sure that the people collaborating on that architecture are the people that are about to implement that architecture. All of the sudden, the ingredients for success are all in the right place at the right time!

Technical Excellence
Technical Excellence can refer to architecture, design, testing, coding, and probably other practices too. Regardless of whether you do BUF, NUF, or Emergent Architecture, the ability to quickly and cheaply extend or change your software architecture at any given moment is directly proportional to your practices around technical excellence. The higher your technical excellence, the more quickly and cheaply you can change direction (<– indeed, this is the definition of agility). Having said that, in BUF approaches, since people make the mistake of thinking that they can “lock down all the requirements and architecture Up Front” , they rarely put in the technical excellence needed for rapid change. Examples of technical excellence are paired programming, collective code ownership, continuous integration, continuous automated testing, build automation and build pipelines, Test Driven Development, Unit Testing, lightweight code reviews, YAGNI, and many other practices. Agile approaches harness change for the benefit of the customer, so don’t forget Agile Manifesto Principle #9: “Continuous attention to technical excellence and good design enhances agility.”

Architectural Runway and the “Last Responsible Moment”
Now, let’s be careful with the “last responsible moment”. Note the word “responsible” in “last responsible moment.” For different architectural pieces, that last responsible moment will be at different times. Some architectural pieces will require a longer runway, and other pieces can do well on a shorter runway. In summary the more complexity and learning in the requirements, people, and technology, the longer the runway needed. If the complexity factors are low, then less runway is needed. Figure out that needed runway length (length of time before that architectural piece needs to be in place), and work your way back to when that runway needs to begin being built.

Let’s look at some examples.

Architectural Runway for a Deployment Platform
For example, determining the deployment platform(the initial UI, logic, and other tiers) for the Minimum Viable Product for a new product probably needs to happen before the first Sprint of developing that product. Having said that, let’s not regress to BAUF, but let’s do execute some upfront thinking and have that deployment platform pretty well figured out before that first Sprint. Since every Sprint has to include at least some small amount of user/business valuable functionality, it’s going to be hard to create some releasable software functionality *and* create your initial deployment platform all in Sprint 1. It is theoretically possible, but not likely.

Architectural Runway for an Open Source Logging Framework (Complexity is low, should be a fairly quick decision)
Another example of “last responsible moment” might be choosing an open source framework to do logging in your system. If logging is forecasted to be in Sprint 23, there is no reason to choose that framework in Sprint 3. You can probably wait until a few sprints before to begin working on that decision.

Architectural Runway for a 3rd Party Processing Component/Framework (Complexity is high, will be a long duration before a decision can be made)
Let’s assume the same as the above — we plan to begin using this component in Sprint 23, and we are currently in Sprint 3. We need to choose a large 3rd party proprietary component that does a large amount of processing or functionality(think accounting, medical, legal, or aerospace). Since it comes from a 3rd party, there is likely going to be a long runway needed to have that choice in place before functionality can be built on top of it. In this example, it might be perfectly fine to start the architectural discussions about which 3rd party framework to use and purchase in Sprint 3. Remember that you’re going to have to leave time for product evaluation, purchasing, legal, training/learning, and maybe even some technical investigation and technical spikes. It could actually take 20 sprints to execute that particular architectural runway and have the architecture first able to support business functionality.

Architectural Runway for logging standards across multiple teams(Complexity is medium, requires time for buy-in from multiple teams)
Remember that logging framework implemented in Sprint 23? Well, we’re now in Sprint 27, and now that the logging framework has been implemented on a couple of teams, the teams realize a need for standardization of use because the framework is beginning to be used widely across the product. At first people didn’t think this was a “high enough concern” to be considered architecture, but then then they realized that the teams were using widely different “logging levels”, that debug logging often crossed module/system boundaries, and that debugging across those modules was getting difficult. The teams finally realized that there would be some major value in ensuring some light standards were followed across the teams. Three might even be a nice wrapper around the chosen logging framework to make it easier for teams to use in a more consistent way. So, in Sprint 27, they began working on an architecture effort to solve those needs. By Sprint 31, the first teams were using the new wrapper and light standards. One thing to note here is that, back in Sprint 20, when they were originally choosing a logging framework, they could have chose to do all of that standardization up front — but that would have been BUF thinking, and they could not have predicted the ways the framework would be used. It’s very possible that no other team would ever pick it up and use it — as happened with most of their open source frameworks. In other words, they would have been heavily speculating needs rather than having a practical, experienced understanding of the real needs. Now, with the experience of a couple of different teams actually using it, in production, they can make better decisions about how to lightly standardize. In this way, they chose to do Emergent Architecture over BUF thinking.

Architectural Runway for Performance and Scalability(Complexity is highly variable here, depends on the product and the target stakeholders for the release)
Hopefully, if you have a new product, you have a Minimum Viable Product(MVP) release that doesn’t require supporting a zillion users concurrently. However, every product is different, so act accordingly. An MVP will certainly need to perform and scale in a way that meets the needs of the initial users. Get this wrong, and you will pay a very heavy price in terms of customer satisfaction and product adoption. As such, for an initial release, this topic is extremely important. The non functional requirements around performance and scalability will often be ordered high in your backlog, as they are both high value and high risk for an initial release. Having said that, sometimes premature optimization is a source of waste. The runway here will also depend on the tools and environments you need to do proper load testing. Choosing and setting those up can often take serious time, so that would extend the runway needed. So, the runway answer here is probably — “it depends.”

Architectural Runway for Government Regulatory Compliance(Complexity is highly variable here, depends on the weight/complexity of compliance, the agility of the agency, etc.)
The guidance for this piece of the architecture is almost identical to the factors involved in dealing with scalability and performance requirements. They are “must haves” that will often be ordered very high in the product backlog due to their risk and value. Note here that we are mainly discussing compliance needed by the product, not the process. (On a related subject, sometimes software process compliance ends up on your Definition of Done, which may or may not affect the product architecture or product backlog directly.)

The above runway discussions are just some examples to think about. The architectural runway for each of these needed pieces of the architecture will be different. As such, the “last responsible moment” to build that runway will require some serious thought as you refine your product backlog. Further, in order to deal with the unknown unknowns, you’d better put a time buffer into that runway since you can’t possible know the unknowable!

In Summary
Hopefully the above has given you a picture of what Emergent Architecture looks like. With each vital part of the architecture, you will need to consider the runway that gives you the “last responsible moment” and some buffer for the unknowns. Because the requirements will be emerging over time, so too will each of the architectural pieces. The architecture will emerge, or evolve, over time, but with intentional forethought to the necessary runway. Remember, due to the number and churn of moving parts (requirements, people, technology, etc) inherent in almost all software development, the *only* way to keep an optimum architecture over time is to give continuous attention to technical excellence. Finally, in order to avoid the gigantic trappings of BUF and NUF, Emergent Architecture is really the only sane choice.

(If you have several teams on the same product or in the same org who need to keep the Architecture of systems excellent, see:  An Agile Architecture Community of Practice

---

Notes:

^1 The focus of this article is Agile Emergent Architecture, so we give only brief attention to a definition of architecture here. It’s worth noting that many thought leaders have multiple credible definitions of architecture. Further, Martin Fowler and Molly Dishman have mentioned that some think the term “architect” is not even the correct metaphor — that maybe “city planner” is the better metaphor.

^2 It’s worth noting that there are some very rare scenarios where a system re-write might not be a sign of failure. For instance, if the system re-write is an early pivot in a product’s lifecycle, it might make great sense and not be due to the above mentioned causes. Having said that, I’ve never personally seen a re-write that was due to a good reason.