Scaling the Practice of Architecture, Conversationally

Architecture need not be a monologue; delivered top-down from the minds and mouths of a centralised few. This article describes another way to do architecture; as a series of conversations, driven by a decentralised and empowering decision-making technique, and supported by four learning and alignment mechanisms: Decision Records, Advisory Forum, Team-sourced Principles, and a Technology Radar

15 December 2021



When “traditional” approaches to architecture break down

I'll be honest, “traditional” approaches to software architecture (i.e. non-coding, decision-taking, diagram-drawing) are hard for me to make work at the best of times. But while using them in the world of continuously delivering autonomous teams I’ve repeatedly found myself faced with an impossible task: to be everywhere, tolerating significant contextual variance, and blocking no-one.

It made me wonder. Was there an alternative?

There was: I stopped taking architectural decisions. Completely.

In this article I’ll introduce this alternative mindset and the associated set of tools and practices which allow me to upend the traditional role of a “Software Architect” while simultaneously bringing the practice of software architecture to the fore across development teams. More importantly, I’ll explain how, within this alternative approach, everyone can do the architecting they need, safely and efficiently, without everything descending into chaos. 1

1: Although there will be a healthy dose of anarchy...

We need more ways to “do” architecture, not less.

The moves in software delivery towards ever-increasing team autonomy have, in my mind at least, heightened the need for more architectural thinking combined with alternative approaches to architectural decision-making.

Ensuring our software teams experience true autonomy raises a key problem: how might a small group of architects feed a significant number of hungry, value-stream-aligned teams? Why? Because in this environment Architects 2 now need to be in many, many more places at once, doing all that traditional “architecture”.

2: And I include myself in this group.

What we need is a workable way to approach the human-scaling challenges of team autonomy and the architectures which manifest as a result.

In the remainder of this article I’ll introduce an alternative way of doing and governing architecture. I’ll explain in detail what it is, how it works, and how you might adopt it yourself. Most importantly, I’ll highlight how to fail, in order that you might succeed.

The most fundamental element: decision-making via the “Advice Process”

Let’s take as our starting point a team which we aim to make maximally independent. Clearly this team will need somehow to engage in architectural thinking and decision-taking, but how?

These “many centres of decision making” are precisely what we need, yet straight away it’s clear that traditional, top-down architecture, with a select group of all-powerful architects taking all the decisions, runs contrary to such a decentralised model. “And yet”, the challenge is voiced “decisions still need to be made - that’s what architecture is”, and these skeptics are right.

These architectural decisions must still be made deliberately - otherwise we’ll be back where we started, or worse. Therefore, the first aspect in this alternative approach, it’s core element in fact, must describe how it delivers on decision-making. It’s called the “Advice Process”.

The Advice Process is the core element of this anarchist, decentralised approach to architecture. It’s greatest quality is it’s remarkably simplicity. It comprises one rule, and one qualifier:

The Rule: anyone can make an architectural decision.

The Qualifier: before making the decision, the decision-taker must consult two groups: The first is everyone who will be meaningfully affected by the decision. The second is people with expertise in the area the decision is being taken.

That’s it. That’s the Advice Process in its entirety.

This apparent straightforwardness hides however a key concept which it’s worth making explicit; while decision-takers are in no way obliged to agree with the advice the folks in these two consulted groups give them, they must seek it out, and they must listen to and record it. We are not looking for consensus here, but we are looking for a broad range of inputs and voices.

A challenge frequently raised against this concerns just how many people must be consulted. It is a valid concern, but a mitigatable one. When deploying this technique we create a checklist to help those in the decision-making seat identify who to speak to, and in which regard. InfoSec impacted? Talk to the CISO. Getting close to PII? Engage Mary in the data team and Vanessa in legal. A potential change to the user onboarding flow? Talk to your UX lead. About to adopt a new cloud service? Chat to Kris the cloud architect. Thinking about a change to your API? Speak to all the leads of the teams who are your consumers.

Sometimes this list of consultees can be a long one. That’s fine. Some decisions are large ones, and the advice-scope is a clear indication of both size and import. Sometimes decisions can be made smaller in scope and many consequently are. Other times the sheer number of folks impacted makes the decide-ee think again. Is this thing which might make their life a little bit easier really worth the effort of consulting all those people? Or, can they split this large decision into multiple, smaller decisions? When decisions do proceed, they are frequently right-sized purely as a matter of expediency. 3

3: The economics of decision-making is dealt with in great detail in The Principles of Product Development Flow by Donald G. Reinertsen. Principles E8: The Principle of Small Decisions, E10: The First Perishability Principle and E11: The Subdivision Principle are particularly interesting in light of what we are discussing here.

Can we push the Advice Process further? Yes we can, and we ought. I always encourage those following it to specifically seek out those who will disagree with them. Freed from the need to agree with what they hear, they inevitably engage far more seriously. Consequently the depth and breadth of advice received is greater. Decisions don’t tend to suffer as a consequence either. Neither does their learning.

Which brings us to the broader topic of benefits of the Advice Process. When deployed, I have always seen better, faster, more accountable decisions, and most importantly decisions which are understood and owned by those who implement them, precisely because the decidee is the one with the need as well as the one who is accountable.

As a side-effect, the pool of available decision-takers also grows, each of whom will soon be on the look-out for decisions which need to be made, and, given the feeling of empowered safety that the Advice Process gives them, flag them up and drive them to conclusion. The fact that a team's need for a decision to be taken can be met by themselves also leads to appropriate levels of bias-to-action, with accountability acting as a brake when it's required.

By working in this way we remove both the need for a fixed and permanent hierarchy and an abiding master decision-taker. It is for these two reasons that the Advice Process is the most fundamental element of this approach to architecture, because decentralised decision-making is the core element of anything which aspires to call itself “anarchistic”.

But wait, did we just remove in one fell swoop all need for we “traditional” architects? Not at all, but clearly our role has changed. In the following sections of this article - which introduces the supporting elements of this approach - we will see a set of rejuvenated practices and tools which allow us to get the teams, and the businesses they underpin, where they need to be.

Before we proceed to these supporting elements it is useful to take a short detour to highlight and discuss the one thing that all the remaining elements to this decentralised approach share, and also have in common with the core element: their focus on conversation, and it’s role in efficiently arriving at, and spreading, shared understanding.

The fundamental role of conversations

Alberto Brandolini, inventor of Event Storming famously quipped “it is the developer’s assumptions which get shipped to production” and he’s right; it’s primarily what a developer understands about a target architecture that matters, not what is in the head or diagrams of a lead architect. This problem is age-old. Eric Evans tackled it in “Domain Driven Design: Tackling Complexity in the Heart of Software” and more recently my colleague Erik Dörnenberg has spoken about it in his presentation “Architecture without Architects”.

To me, it is this architecture, the one which is in the heads of those writing the code, that is the most important. In adopting this decentralised approach, where the practice of architectural decision-making is much more dispersed, this problem is in many ways, mitigated.

That’s something which helps me sleep at night.

However this decentralised, anarchistic approach then places front and centre another problem which all architectures must address: the delivery of a coherent whole. Here, we would seem to be at a disadvantage with our alternative approach. If everyone is empowered to make a decision, how do we, the “traditional” architects and ones who perhaps care most about the overall end result, ensure that the sum of all the individual decisions combine to form a coherent whole? How might we incorporate a longer-term perspective into those same decisions? And how might we support those who suddenly find themselves taking on levels of responsibility for which they might not feel comfortable?

Fortunately, another practitioner, and thinker in this space, Ruth Malan, has seen this before and shares the answer in her article “Do we still need architects?”:

[In order for an architecture to be successful] it is very much about ensuring that conversations that are needed to be happening are happening - not always initiating them, nor always helping to focus or navigate them, but ensuring they do happen […] and guiding when needed

-- Ruth Malan

Our adoption of the Advice Process opened up the space for anyone to make decisions, but it has also put conversations, the responsibility to seek out expertise, and think about impact at the core. The remainder of the elements of this approach, each of which supports the core element focus specifically on ensuring those conversations are as timely, focused and effective as possible. There are four of them:

  1. a thinking and recording tool
  2. a time and place for conversations;
  3. a light to illuminate and guide towards a unified direction;
  4. a means to sense the current technical landscape and climate.

We'll cover each of these in turn in a few seconds, but first I need to clarify two things.

What about strategy and cross-functional requirements?

It's worth a few sentences on what isn't covered in this approach: technical strategy and cross-functional requirements (CFRs).

Clearly both are essential for all software endeavors of any meaningful size:

A well evangelized strategy can help the organization advance by having decentralized teams prioritize technical activities which are best aligned with the org's maturity and needs. Clearly the best technical decisions are those which support the strategy, and when this is the case this can be called out clearly.

A clear set of testable CFRs also helps a decentralised set of teams ensure that they look beyond their immediate, local delivery, and meet the minimum requirements for playing coherently in the shared ecosystem.

However, technical governance refers to both these rather than encompassing them - they contribute to the context within which it operates - and so I've not gone into any detail on them here. But what does governance include beyond a means to ensure good technical decisions? Let's take a look.

The Four Supporting Elements

1. A thinking and recording tool: Decision Records

The first supporting element is Architectural Decision Records or ADRs. These are lightweight documents, frequently stored in source code repositories alongside the artefacts they describe. Now, there are a variety of formats which various adopters have chosen to champion, but the key elements which I insist on are as follows: 4

4: There are actually two more, but we’ll get to those and discuss them separately in detail in supporting elements 3 and 4.

Elements of an ADR
namedescription
titlewhich includes a unique identifier, and the decision itself (e.g. “ADR001 - Use AKS for Kubernetes Pods”)
statustypically “Draft”, “Proposed”, “Adopted”, “Superseded” and “Retired”
decisionthe decision that has been taken in a few sentences (frequently bold or italicized so it stands out)
contextthe forces and current contextual circumstances which have necessitated this decision
options consideredeach option considered, described briefly, with pros and cons. (Typically the option proposed / adopted comes first in this list)
consequencesthe ramifications of this decision, both positive and negative
advicethis reflects the raw outputs from following the Advice Process. It is here that all advice given is recorded. This ought to include the name of the advice giver, and the date the advice was given. This can frequently take the forms of comments, and if these are provided directly by the advice-giver, then recording the meta-data is automatic.

I’ve found in practice that having such a lightweight ADR template structure is not only a great way to record architectural decisions - it also helps teams learn to make architectural decisions. These key elements operate like a thinking checklist, and prompt the decide-ee regarding what they need to think about, and more importantly have conversations about.

What’s more, ADRs serve to reinforce the Advice Process by making it a requirement on ADR authors to capture and record all advice they get. I also encourage authors to engage with this advice directly in their ADR options section, whether they choose to follow it or not. It is one thing to seek advice and write it down. It is quite another to actively wrestle with it. The fruits get sweeter the more you engage.

It will come as no surprise to learn that consequently, a series of ADRs, and their surrounding conversations provide an excellent learning ground for people wanting to begin to take on the task of decision-taking; everything is out in the open, including the dissent and compromise-making. Less experienced practitioners of architecture can peruse the history of what went before them quickly and easily, see good (and quite likely less-good) examples, and see decisions being taken (and perhaps also being revoked when circumstances change / the team learned more). They are almost a thinking and decision lore for a set of software, written in the hand of those who contributed most to it.

While sadly I can’t share with you examples of these conversations I’ve had with my clients, there are some great examples of ADRs out on the public internet, courtesy of Thoughtworks-alumni Wisen Tanasa and his startup Upmo. I’d encourage you to take a look. They come blessed by none other than Michael Nygard himself.

2. A time and place for conversations: The Architecture Advisory Forum

The second supporting element in this alternative approach exists to make all the conversations supporting this advice-seeking easier: a weekly, hour-long Architecture Advisory Forum (“AAF”).

Fundamentally, this is a regular and recurring place and time for conversations. Your ideal attendees are delegates from each team as well as your key representatives from your Advice Process checklist. However, the invite should remain completely open to encourage transparency and openness. The timeliness and quality of the conversations which take place is a key indicator of success, but equally important is the breadth and diversity of views shared, and the same goes for the contributors. If architecture is being “done” here, and lessons shared and learned, then you’re winning.

The standing agenda typically begins as follows:

  • team representatives quickly share new spikes (giving early warning of probable future decisions and allow the attendees to share existing knowledge and experience)
  • discussions about each new “proposed” decision (presented by those making the decision, captured ahead of time in the form of an ADR)
  • a re-visit of other decision-statuses (we timebox these, both to limit the window for incoming advice, and also to allow us to revisit a decision which we made with imperfect information)
  • a look at our collective four key metrics, our cloud spend trends, and finally
  • any other business (aka “AOB”)

A cursory glance might give the impression that an AAF is just a new title for a standard meeting. The one typically known as a “Tech Advisory Board”, “Architecture Decision Forum” or “Architecture Review Board”. There are however several key differences.

Firstly, the Advice Process reigns. Decisions taken to the AAF are still owned and made by the originators. The only thing other attendees can do is offer advice, or suggest additional people to seek advice from. Hence the name.

This brings us to the second key difference. Given the Advice Process qualifiers, the invitees to the AAF are those typically affected / possessing relevant expertise. This means those typically present include representatives from each feature team (and not just the lead; BAs/POs and QAs are frequently present), people from other programmes of work, UX, Product, Operations, and occasionally senior execs.

The combination of these two differences leads us to the third, and most important key difference: the conversations. The Advice Process is great, but it’s conversations can frequently be 1-1. When they take place in an AAF there is an audience, so many people can listen and everyone can learn. The amount of organisational, domain, legacy, and experiential information and architectural skill-deployment shared at these sessions is unlike anything I have ever seen, and despite being a potentially dry meeting, it is the most well-attended, and most broadly participated hour of our week. It is one of the most significant contributors to the quest for a learning organisation that there is. AAFs encourage disagreement, and celebrate failure / changes of decision based on lessons learned. This all combines to broaden and deepen the general understanding of an architecture, virtually guaranteeing it ends up in the running software.

3. A light to illuminate a unified goal: Team-sourced Architectural Principles

Having architectural principles is not new, though sadly I rarely encounter serviceable ones. Always important, in a world of highly-autonomous-teams they become essential because they are the means by which an aligned delivery direction is achieved without the need for control.

So what makes a good architectural principle? Firstly, it must provide a criteria with which to evaluate our architectural decisions (which in practice means it must be specific, measurable, achievable, realistic and testable, aka “S.M.A.R.T”). Secondly, it must support the business’s strategic goals. Thirdly, it must articulate the consequences / implications it necessarily contains within it. Finally, taken together as a set, they should number neither too few to cover the key needs which architectural principles meet, nor too many that teams cannot remember them all.

There is a great deal I could write here about bad architectural principles but I’ll stick to the key aspects. Firstly, they are not practices. Practices are how you go about something, such as following TDD, or Trunk Based Delivery, or Pair Programming. This is not to say that practices are bad (indeed Dr Forsgren’s “Accelerate” is full of recommendations regarding for their set of circumstances) they’re just not architectural principles.

Watch out for slipping into the other end of the scale too - general principles. “Keep it simple” and “Don’t repeat yourself” are principles, but they’re not architectural. Nor are the various principles you’ll see around project planning, and software quality management. What we need are means to direct and evaluate our architectural practice and decisions. What we need is something which helps me pick between various approaches to implementing micro-frontends, or helps me decide if it really makes sense to hand-roll my own OAuth 2.0 implementation, or guides me in evaluating self-hosted Lucene on AWS vs Amazon Elastic Search Service.

Given all this, now let’s share a good principle, based around the Team Topologies “Stream-Aligned Team” organisation model:

Title: Value independence of teams most highly

Subtitle: Split solutions along team lines

Rationale: The strength of our approach to building and running our products relies fundamentally on the independence of our teams. The downsides to this are acknowledged, but the upsides are felt to outweigh it, especially when the difficulty of predicting future needs is taken into consideration.

Implications:

  • Duplication of both function, and data, will inevitably arise. Rather then fight this, we embrace it, acknowledging the need, in certain circumstances, for noticeable eventual consistency and data replication
  • The combined licencing, runtime and support costs of multiple third-party solutions may be higher than the costs of a single, shared, cross-product-team solution
  • Solutions can be designed for the needs of the team which owns and runs them. They need not concern themselves the needs of other teams
  • Both systems and the third-party services / solutions they are build on will tend to be smaller, and more specific-task-focussed
  • Teams who go their own way need to self-support any third-party services / solutions which they adopt independently

If you want to see more examples, please have a look at the publicly-available John Lewis “Software Engineering Principles”. 5

5: Please note, at the time of writing, the John Lewis principles includes a few items which fall outside our definition of an “architectural” principle. I’m thinking of for example “Understandability” and “Performance Importance”. This doesn’t mean these aren’t bad goals, they just don’t meet our definition.

So far, so general. Nothing I have said so far in this section would be controversial in any approach to architecture. Why then am I stressing these points so much? Not only is the importance of architectural principles heightened in this decentralised approach, but also everyone concerned needs to know how to structure them and what good looks like because they will be sourced from, and maintained by, the teams themselves.

Our approach is, to a great extent, taken directly from the excellent “The Art of Scalability” by Abbot and Fisher. While their book assumes a slightly more top-down, hierarchical approach to architecture than presented here, the authors very much recognise the impact of the human element on their topic. In fact, the edition I read had been significantly re-written to give more weight to this perspective. One key aspect of this is their argument that for any architectural principle to be successful, teams which deliver against them need to feel a sense of ownership over it.

I’d encourage you to take a look at their book for a wealth of detail on how to source these from the collective. Suffice it to say, when presented with the business’s strategic goals, the “S.M.A.R.T.” criteria from earlier, and a broad set of invitees from across technology and beyond (yet again your AAF invitee list will prove invaluable here) you will rapidly and collectively arrive at 8-15 principles which will serve you well. 6 It’s well worth capturing the adoption of a principle as an ADR. These will be very lightweight (don’t fall into the trap of repeating the principle itself) offering a great opportunity to articulate why this principle is important.

6: It’s worth pointing out that, while their JADs and ARB’s in the following chapter don’t fit so well the approach we’re following here, they do contain some great thoughts on when to revisit / update the principles, thereby ensuring they keep fresh through constant usage and re-evaluation.

There is one final point to make on the principles. Remember that this approach is aimed at supporting team autonomy, so one key role played by our principles is as a minimal viable set of understandings and agreements between everyone. This raises a key point, because one thing we ask teams to explicitly flag in their ADRs is not just the principles which apply, but also when their decision conflicts with one or more principles. This becomes a great point to engage the Advice Process and the power of the collective at the AAF to really get all the best minds and varied perspectives on the problem, and then record all this in the ADR. Yet again, the various elements support each other, amplifying their benefits, and helping us get to successful architectures. Remember, if, as a consequence of this a principle changes, call that out as a separate ADR that supersedes the original.

That’s architectural principles covered, which play the role of a guiding light for everyone to aim for, but how do we also take note of our surrounding landscape and climate? Architectural decisions are also frequently based on what everyone else is doing, who has which skills, and what the general trends in the tech industry are. Enter the fourth and final supporting element: your own Technology Radar.

4. A tech landscape and current climate sensing tool - Your own Tech Radar

Many people have heard of the ThoughtWorks Technology Radar - an opinionated guide to current trends (predicted, current, and receding) in software languages and frameworks, tools, platforms, and techniques. It’s strengths lie in how it visually represents both the current landscape and the movements of various “blips” across it, allowing viewers to very rapidly see (for example) what is up and coming in the world of front-end frameworks, what’s current flavour of the month, and what’s beginning to fade.

Sadly, far fewer know about the fact you can build your own radar. The “BYOR” allows you, as a collective, to capture and map out your local version of the technology trends you see across your organisation. It’s very configurable too. In my most recent usage we kept the quadrants (Techniques, Tools, Platforms and Languages & Frameworks) but changed the rings to reflect the transit of technologies through our programme of work (they became “experiment”, “adopt”, “hold” and finally “retire”).

As with the architectural principles, these radar blips need to be crowd-sourced in a workshop. The first run of this will capture everything you have now in your organisation - a baseline sweep or scan if you will. Prior to this you need to figure out how wide you will go (org-wide? just your project? Will you include disciplines such as Ops and UX? etc.) and what your quadrants will be. It’s also possible to add extra fields for data capture, but I typically try to keep it simple. The first time you do this it can take a fair time (we’ve taken four hours and more before) but this is because it is essential that you involve all team members, not just architects, and the end result will give a great overview of the landscape and prevailing climate, and brings many discussions about where effort should be directed, and where it should be reduced. And just as with the principles, give rise to a general aligning of team understanding.

What about the usage of your radar? As with the principles, there is also a place in our ADRs for “Relevant Radar Blips”. This is where we flag both adherence to the existing landscape as reflected in the current radar, but also, and more importantly, potential changes to the existing radar which this decision will introduce. Perhaps it’s the spiking of a new framework, or a move from “experiment” to “adopt” for a specific practice.

Again, this is great grist for the AAF discussion forum, and great content to capture in the ADR itself. You can even go so far as linking specific types of blip appearances and movements to the need to submit ADRs, though in my experience this happens anyway without anyone having to push it explicitly. Remember, your goals here are the broadest engagement with your evolving architecture as possible, as well as a growing architectural mindset across all team members.

How about keeping your radar up to date? I've seen quarterly cadences work, and half-yearly too. The key is to pay attention to how the radar is being consumed (or not) at the AAF and elsewhere. That should give you a good idea when it's worth investing in a refresh.

How this typically works in practice

Given all this, how might you see it all working in practice? Let’s take a look…

When the need for an architectural decision first arises it will most likely be vague and possibly poorly understood. It’s therefore great to open up a new ADR template right away and start trying to fill it in.

First to be tackled is the “context” section. To attempt this we need to understand the “why” of our decision as well as surrounding forces which we need to balance. We’ll probably rapidly realise we need to do some research to be able to complete even this short section.

Early ports of call in this research ought to be the architectural principles and radar. The principles, you’ll recall, give us an idea of the direction of travel which our ideal solution will ideally manifest. Not all will be relevant, but some principles ought to help our decision-making. Recall that it’s an architectural principal’s primary goal to assist in the evaluation of multiple technical possibilities, and highlight the one which fits best.

Sometimes, the experience will be a little different. One alternative is that a relevant principle cannot help you pick between two options, then there is either very little difference to choose between, or potentially, your principle isn’t S.M.A.R.T. enough. This is a good reason to revisit the principle and re-define it.

The other alternative also may end, perhaps a little later, in a re-evaluated set of principles. These arise when in order to make your decision you feel inclined to contravene one or more principles. That’s ok; good decisions can go against the principles, but to do so you’ll need to state clearly why this course of action was the right step to take. Overriding a principle is a significant step because it means we are effectively diverging from the general direction of travel. Consequently, the decision, and resulting ADR must be clearly argued and strongly justified. It might also signal the time to revisit the principle in light of this development.

The radar in comparison is a lot more advisory in nature. It will give an idea of what, if anything, is the current de-facto standard in our problem space, what’s been done in the past, and what other teams might be experimenting with. Going a different way is a lot less likely to raise eyebrows, but it is again a definite reason to address the deviation in the ADR.

Given all this, we can start to come up with our key criteria to evaluate options against, as well as a list of alternatives. Perhaps we’ve realised we really need to do some homework and in which case we might spin off a timeboxed Spike to learn more about something. Our ADR thinking will help write super-clear acceptance criteria here. If we don’t need to do a Spike, we will think instead about seeking advice.

Having these inputs from the context (which will include, for example, a technical strategy), applicable principles, radar blips, and key criteria / alternatives will in turn help us think about who to speak to for advice. In my experience it is helpful to engage in this when you are relatively confident you understand the problem space / need, but before you become too attached to a specific solution. When you go and seek advice, spend most time and effort speaking to people who will disagree with you; those who you know think along different lines and where you know you will have blind spots. Not only this, challenge yourself. Consider “what’s bad about this alternative? What are its shortcomings?” Spend the most time thinking about the alternatives which challenge your decision most directly and fundamentally.

Before you have the discussion, it helps to have the ADR ready in a rough form, and share it with them in advance. This will give them thinking time. Then when you do meet, go through every element of the ADR template. Have you missed something from the context? From the principles / blips? From the evaluation criteria? Solicit and capture their advice on all these things. More importantly, ask them why they are making these suggestions. It is the answers to these questions that the keys to learning to make better architectural decisions lie. Your advisors will help you understand how they see problems, what they’ve encountered that was similar in the past, and even entire aspects which you don’t even think about. 7

7: Classic examples include people who’ve had to find, hire and retain developers and people who’ve been the “support” end of a decision thinking differently about new technologies.

What about the AAF? Isn’t that the place to gather advice? Yes, and everything I’ve shared above should guide you wherever you have the conversation, but for the first few decisions a team or individual takes it really helps to have these in a more targeted way with the key advisors. The ADR you present before the AAF will then be really solid and focused before it is shared widely, meaning that the resulting AAF conversations will be richer and more focused. Sometimes a quick conversation in the AAF will lead to a subsequent, more in depth, 1-1 conversation.

Remember, once you have advice, wherever it came from, you must roll it into your ADR. You need not take the advice given, but you have to record it. An excellent practice here is to prioritise things in your writing which people would find non-intuitive or surprising. If you disagree with key advice, state how and why. If you are doing something new, make it clear why the current way doesn’t work for you. Remember to use the principles. Sometimes they will support your decision. Make it clear how. Sometimes you will have to contravent them. Make it clear why. Your goal is to make the reader understand why you took the decision you took. If you meet this goal, then you will not just have a solid decision, you’ll most likely have learned a great deal in the process. 8

8: There are some great blog posts which meet this criteria that are in many ways just extended ADRs. “Why not Rust?” and “No, we don’t use Kubernetes” are excellent examples, and worth reading if you want to see some really great decision-taking (and cargo-cult-avoiding) in progress.

Before I close this section, remember, all decisions are point-in-time, and no-one can ever foresee every eventuality, but you want to be able to go back later and still feel good about a decision, given what you knew / understood at the time. In this, the context and criteria you capture are key. This re-visiting of decisions is another great learning tool. With the benefit of hindsight you can now ask questions such as: “did we understand the context well enough?” and “were we honest enough with ourselves about what we knew and what we didn’t?”

How to fail

So that’s the extent of our alternative, decentralised, anarchistic approach to architecture, and an idea of how it typically all fits together. But before we conclude, we ought to address one final aspect - the key ways in which you can fail. Let’s enumerate them.

The majority of failures you will see will actually be good ones - mini failures as decisions are taken by those who are less experienced. These are good because the process facilitates quick decisions, by those who need them, and more importantly, it facilitates transparency and rapid identification of failures (as those who took the decision will be aware of the issues as they code it) and a safe means to re-visit, and share the learnings. Embrace these, calling them out specifically and celebrating them in the AAF. This is a key aspect of building a learning culture.

To learn most effectively you need to feel safe, and when learning collectively everyone benefits from the broadest, most diverse range of inputs contributing to discussions. Remember, in this approach, we are explicitly not looking for consensus, but we are looking for a broad range of inputs and voices. It is here that the next failure mode lies, and it is far more insidious and damaging than the first. This second failure mode arises when you, in your job as a conversation-starter and space-holder, fail to include all who ought to be contributing and deciding and learning. For a great part early adoption stages of this style of architecture can feel like great successes. “It’s working! More and more people are taking decisions, writing them up in ADRs, giving advice and discussing them in the AAF! I’ve never seen such engagement with principles before!”. Only later, on reflection, will you realise the gains could have been far greater. It is precisely when this first flush of satisfaction strikes that you must be most on your guard. Are you really observing mass participation and learning, or is it a core group of usual suspects? You mitigate this problem actively. Watch out for who contributes. Amplify voices and ensure others listen to the quieter contributors. Make sure influence is balanced and not based on reputation, tenure or place in the hierarchy. Actively encourage many viewpoints and highlight the value it brings so that it becomes self-sustaining.

The third failure mode is an early warning that you are encountering the preceding failure mode, however, this one lives more in the grey area between desired and unwelcome. As you proceed along this journey you will uncover off-the-grid decisions. Decisions which never came up at the AAF, and which never made it into an ADR. There are two ways to approach them. The first, correct way, is to treat the discovery as what you hope it is - an honest mistake, and an opportunity to learn and teach others. Perhaps the decide-ees weren’t even aware of the fact it was a key decision they were taking. Perhaps they were under pressure from elsewhere. Perhaps they thought it wasn’t as significant as it turned out to be. Perhaps they felt they would be shouted down in the AAF. Whatever the reason, treat it as a way for both them, and you, to learn. To improve the process. The other, wrong way to treat these, is to fall back to old ways, and take back control. Which takes us nicely to the failure mode which completely destroys this approach and all that it promises.

It’s easy to slip into this fourth and most dangerous failure mode, and so it needs constant vigilance on your part. The only thing which needs to happen to trigger this is for “capital-A” architects such as yourself to fail to trust people; it is to not practice what you preach; it is to not clear enough space for the mini failures and consequent learning opportunities just mentioned; it is to continue to perform “shadow architecture” behind the scenes to make sure things still go how you think they ought to, despite all the signals from elsewhere. The sole benefit of this failure mode is that it becomes evident very rapidly as all the benefits I’ve listed above fail to materialize.

If you’re wondering if it’s this key failure mode which makes this approach to architecture hard to pull off you’d be right. I have been lucky in the past. Colleagues have called me out when I’ve made decisions for others, and I’ve caught myself getting frustrated that folks don’t know what I know. But then I realise I’m failing in my real task as a practitioner of architecture - I’m failing to get the right conversations happening, with the right people, at the right time. Remember that (perhaps even task others with calling you out when you fail to stick to the process) and you’ll be surprised how easy (and satisfying) it is to succeed.

The five elements again, and possible further steps

Given we now have both the good and the (potentially) bad, as well as a collection of failure modes to look out for, let’s conclude. We recall we have the five elements of our alternative approach to architecture:

One core element: Advice Process

Four supporting elements:

  • Architecture Advisory Forum
  • Lightweight ADRs
  • Team-sourced Principles
  • Your own Tech Radar

Hopefully I’ve made it clear that while none of the elements may be new to you (aside from perhaps the Advice Process) there is something very different. This difference lies in the interplay/mutual reinforcement between all of these against a backdrop of conversations, learning and safety. What is hopefully enticing is the fact that, in my experience at least, this is far more likely to provide successfully deployed architectures, now, and into the future. It is my go-to for scaling myself, and making sure that the teams I work with deliver on the promises we have made to our users, which is, after all, the goal.

Afterword: What next?

Of course, while I’ve tried to keep this article focused, it’s possible to go further. I’ve seen teams working in this way naturally starting to pave their own roads - self-serving their own delivery platform before a (Delivery) Platform Team comes into existence. (See my colleague Evan Bottcher’s excellent “What I talk about when I talk about Platforms” and Skelton and Pais’ Team Topologies for more detail on this.)

I’ve also seen teams put in place their own Architectural Fitness Functions (and not just around run cost) so that they know when the collective architecture strays outside its intended bounds.

The biggest lesson I’ve learned is that, once you empower people, giving them an environment in which to succeed, and recognise their successes, they will rapidly, and as a collective, start thinking about things which haven’t even crossed your mind. That’s the real benefit of this kind of approach: access to the collective intelligence of the many, over reliance on the much more restricted intelligence of the few.


Footnotes

1: Although there will be a healthy dose of anarchy...

2: And I include myself in this group.

3: The economics of decision-making is dealt with in great detail in The Principles of Product Development Flow by Donald G. Reinertsen. Principles E8: The Principle of Small Decisions, E10: The First Perishability Principle and E11: The Subdivision Principle are particularly interesting in light of what we are discussing here.

4: There are actually two more, but we’ll get to those and discuss them separately in detail in supporting elements 3 and 4.

5: Please note, at the time of writing, the John Lewis principles includes a few items which fall outside our definition of an “architectural” principle. I’m thinking of for example “Understandability” and “Performance Importance”. This doesn’t mean these aren’t bad goals, they just don’t meet our definition.

6: It’s worth pointing out that, while their JADs and ARB’s in the following chapter don’t fit so well the approach we’re following here, they do contain some great thoughts on when to revisit / update the principles, thereby ensuring they keep fresh through constant usage and re-evaluation.

7: Classic examples include people who’ve had to find, hire and retain developers and people who’ve been the “support” end of a decision thinking differently about new technologies.

8: There are some great blog posts which meet this criteria that are in many ways just extended ADRs. “Why not Rust?” and “No, we don’t use Kubernetes” are excellent examples, and worth reading if you want to see some really great decision-taking (and cargo-cult-avoiding) in progress.

Acknowledgments

Many thanks to all those who helped me to get this post to state it is now: Martin Fowler, Nimisha Asthagiri, Nick Robinson, Rob Horn, Ian Cartwright, Tim Cochran, Carl Nygard, Marty Abbot, Mel Mitchell, Pete Hunter and James Brown. Thanks for all your brain-power and inputs.

Significant Revisions

15 December 2021: Finished publication

13 December 2021: Published Tech Radar

09 December 2021: Published Architectural Principles

08 December 2021: Published Architectural Advisory Forum

01 December 2021: Published Decision Records

30 November 2021: Published The Advice Process