From Engineer to Architect

Pragmatism
What Matters Most, Right Now?

Article 4 of 9 The Craft of the Decision · 9 min

Someone asks whether we can build the thing. There are only three honest answers, and I have watched architects lose credibility by giving all three in the same tone of voice. "No", "not now", and "let's do it" are three completely different mindsets. Each needs a different mechanism of thinking behind it, a different kind of evidence, and a different conversation to make it stick.

// the crux

Perfect architecture does not exist. Right-fit architecture does.

// in one breath
  • Resilience, scalability, availability, performance, cost, compliance. All of them matter, and none of them can be first.
  • The three answers an architect gives are not three volumes of the same answer. They are three separate machines, and using the wrong one is how trust leaks.
  • Where the paralysis actually comes from, the four tools that each answer a different question, and what pragmatic architects get known for.
↳ the whole value system · From Engineer to Architect – the fourteen mental models this series is built on, and where this one sits.
the impossible list

You Cannot Optimise Everything At Once

In software and cloud architecture we juggle the same handful of demands on every project: resilience, scalability, availability, performance, cost, compliance, and the delivery date somebody already promised. Every one of them is legitimate. Every one has a person behind it who is right to care about it. And they do not fit together at their maximums, which is not a failure of engineering. It is arithmetic.

So the architect's strength is not knowing all of them equally. It is evaluating what matters most in this moment, for this context, for this system, for this business objective, and being able to say which one it is without flinching when someone pushes back. Across systems in several domains and three countries, that judgment is the thing I have seen separate architects people trust from architects people merely consult.

Sometimes the answer is uptime, because a minute of downtime has a price somebody can quote you. Sometimes it is cost, and I have been in the version of that story where the bill mattered more than the elegance, which is its own essay. Sometimes it is compliance, because a regulator does not care how clean your diagram is. And sometimes, honestly, it is just getting a working version into production so the business can find out whether anyone wants it.

A dial with a bronze needle pointing at a single lit segment marked this quarter, pick one, beside the competing priorities of resilience, scalability, availability, performance, cost, and compliance, under the line all of them matter, none of them can be first.
The dial points one way at a time. Naming which way, out loud, is most of the job.
three answers, three machines

"No", "Not Now", and "Let's Do It" Are Not the Same Skill

This is the part I wish someone had told me early. These three answers feel like points on one scale, from most negative to most positive. They are not. Each is a separate decision, resting on a separate kind of evidence, and each fails in its own particular way when you reach for the wrong one because it was socially easier in the moment.

"No"
a boundary
// what it really isA judgment that this conflicts with something that is not negotiable: a regulatory line, a security boundary, a commitment already made to someone else.
// what it needs from youThe specific constraint, named out loud, and an alternative that gets them most of what they wanted. A "no" without a door in it is just an obstacle.
// how it failsSaid too often, or said from taste rather than constraint, and people stop bringing you things early. They bring them once the decision is already made.
"Not now"
a sequencing call
// what it really isAgreement that it is worth doing, plus a claim that something else is worth doing first. This is the hardest of the three and the one most often faked.
// what it needs from youA named condition and a date. "Not until we are past the migration, so let's look again in October." Without those two, it is not a decision.
// how it failsIt becomes a polite "no" that nobody has the courage to say plainly, and the backlog fills with things everyone has quietly agreed will never happen.
"Let's do it"
a commitment
// what it really isAn agreement to spend real capacity, and to accept the specific risk you have already looked at rather than the one you have not.
// what it needs from youWhat you are consciously trading away to make room, said in public. Every yes is a no to something else, and if you do not name it, someone else will discover it.
// how it failsEnthusiasm without capacity. The team hears commitment, the roadmap never moves to reflect it, and the yes quietly becomes an overrun.

Run your last month through those three columns. In my experience the honest audit is uncomfortable, because most of us have a favourite. Some architects say "no" in a tone that means "not now" and leave people confused about whether the door is open. Others say "let's do it" because the room feels good and the capacity conversation happens later, alone, with the team who has to absorb it. Matching the mechanism to the answer is most of what people mean when they call someone reliable.

right-fit, in practice

Pragmatism Is Understanding Trade-offs Without Being Paralysed By Them

There is a failure mode that looks exactly like rigour from the outside. You can see every trade-off clearly, hold all of them in your head at once, and therefore never commit, because every option has a cost you can articulate. Analysis becomes a very sophisticated way of not deciding, and the team ships nothing while you are being thorough. The previous essay is about seeing the trade-offs properly. This one is about the discipline of then actually choosing, on time, with what you know.

Right-fit is what that looks like in practice. It means knowing when to run Kubernetes yourself and when a managed control plane like EKS is just fine. It usually means the boring option is correct, and being senior enough to say so without feeling like you are underselling yourself.

A scheduled job, once a night
A cron job and a log file. An event-driven pipeline for something that runs on a timer, with no consumer waiting on it, is architecture bought for a problem you do not have. I have seen the simple version survive years longer than the sophisticated one it replaced.
One service, one team, modest load
A managed platform that takes the operating burden away. Container orchestration is the right answer often enough that reaching for it becomes reflex, and the honest question is not which orchestrator, it is whether this workload has earned an orchestrator at all.
Many services, real scale, a platform team to run it
Now the orchestration earns its keep, and the managed control plane earns it further, because the hours you save are hours nobody has to spend patching. The tool did not change. The context did.
Nobody has validated the idea yet
The smallest thing that can go to production and be measured. Not because quality does not matter, but because the most expensive architecture is a beautiful one built for a product nobody wanted.
// pattern to refuse

Architecture bought on reputation

The most expensive decisions I have watched were rarely wrong in principle. They were right for somebody else's scale. A pattern gets adopted because a company a hundred times your size wrote a good engineering blog post about it, and the cost of running it lands on a team of six who never had that problem. Borrowed architecture arrives with the other organisation's constraints baked in, and those constraints are the part nobody copies deliberately.

the instruments

Four Tools, Four Different Questions

Structure is what stops "align with the business" from being a slogan. But a tool used on the wrong question produces a confident answer to something nobody asked, so it is worth being precise about which instrument answers what.

Decision matrix
"Which option, given what matters here?" The weighted scorecard. It is the workhorse, and it is the one that makes the ranking visible to everyone in the room. Worked example in the previous essay.
Architecture scorecard
"How is what we already run actually doing?" Not a choice between options, a standing health read across resilience, cost, security, and operability, so the next decision starts from evidence instead of anecdote.
SWOT
"What are we walking into?" Best before the options exist, when the question is still shapeless. Its real value is the two quadrants teams skip: the weaknesses you own and the threats you would rather not write down.
Eisenhower matrix
"What deserves me this week?" Urgent against important, and the reason it belongs in an architecture essay is that the important-but-not-urgent quadrant is where every piece of architectural debt is born and quietly funded.
// the rule that saves the most time

Timebox the ideation, not just the delivery

Give the exploration a deadline in the same sentence you give the problem: two days to compare three options, decide on Thursday. An open-ended design phase does not converge on its own, because there is always one more option worth a look and one more benchmark worth running. The timebox is what converts thinking into a decision, and action taken on Thursday with eighty percent of the picture beats perfection that arrives in November.

// take this into Monday
  1. Name the one dimension that wins this quarter, in writing, where the team can see it. Uptime, or cost, or compliance, or speed to market. One.
  2. Say what you are trading for it. An unnamed sacrifice gets discovered later by whoever is on call.
  3. Audit your last three answers. Were they really "no", "not now", and "let's do it", or were two of them the same answer wearing different clothes?
  4. Put a date on every "not now". A condition and a calendar entry. That is the whole difference between sequencing and stalling.
  5. Leave one question open on purpose, and say that you are doing it. "We do not know yet, and here is when we will" is a legitimate architectural position, and pretending otherwise is how teams end up committed to a guess.
what it earns

What Pragmatic Architects Become Known For

This discipline compounds in a way that is easy to miss year to year and obvious across a decade. These are the things that actually accumulate.

The trust of the business
Stakeholders stop treating architecture as the department that says no, because your answers track their constraints rather than your preferences.
Systems that evolve
Right-fit choices leave room to change. Systems built for an imagined future tend to be rigid in exactly the places the real future arrives.
Debt that stays visible
Pragmatism is not permission to accumulate quietly. It is taking debt deliberately, writing down what you took, and keeping it on a list somebody actually reviews.
Comfort with the open question
You stop needing an answer for everything in the room. Leaving a question open and named is stronger than closing it with a guess that hardens into a decision.
Momentum, not just mastery
Deep knowledge earns respect. Delivering steadily, with judgment, is what gets you invited into the decision before it has been made.
owned publicly
// I believe this

Being practical is not being average. It is being outcome-focused, user-aware, and always grounded.

The question is never "what is the best architecture". It is "what matters most, right now, for this system and the people paying for it". Ask that first and the options sort themselves, the arguments get shorter, and the decision you make on Thursday is one you can still defend in October. Perfect was never on the table. Right-fit always is.
// carry forward

Choosing what matters most assumes you have seen what could go wrong. Article 5 turns to inverted thinking: asking what could fail before what could work, and writing down the failure factors while there is still time to design around them.