Pages

Sunday, November 12, 2023

Agile vs Waterfall - the wrong question

22 years after the "Manifesto for Agile Software Development," the battle for Agile vs. Waterfall still isn't settled - despite many prominent publications, such as a recent HBR article, trying to "settle the debate."

Unfortunately, this debate can't be settled - because it starts with the wrong question. We will take a look at the question that needs to be asked.

Finding the right question

Most authors start with a question that doesn't bring us any closer to a sensible answer, quite often out of ignorance on what "Agile" is.

What "Waterfall" isn't.

Let's examine some definitions you may find floating around.

"Everything that's wrong"

Well, congratulations: with that definition, even adding too much salt to your breakfast eggs is Waterfall. Fun facts - even "Agile" doesn't prevent things from going wrong, and not even "Waterfall" projects always go wrong (would that mean that Waterfall isn't Waterfall because Waterfa ... argh, paradox!)

A sequential, linear approach from idea to completion.

That's crudely based on Winston Royce's paper from the early 70's - but face it: no company actually works like that, and even Royce wasn't suggesting that. Even "Waterfall," in industry common practice, is a continuous, incremental development approach. Just with very big increments, and extremely slow feedback cycles (it's not unrealistic that it could take 2-5 years between when a requirement is defined until people get the thing they actually need.)

A stage-gated approach to product development.

That definition has two fundamental flaws:
First, you can stage-gate "Agile" approaches as well, and before 2006 (i.e. five years after "Agile") - stage gating wasn't even a thing in most enterprises: You can well do non-Agile projects without stage gates.

You can find another, much more comprehensive article on other things that "Waterfall" is not referenced here.

So we're in a bit of a pickle: We can't even find a non-strawman definition of "Waterfall" that would go beyond the completely fictional definition used in Royce's paper for how to not do things.

What "Agile" isn't.

Let's do the same favor to "Agile" and pull out some of the common misleading beliefs floating around:

Scrum is not Agile

Hard to overemphasize this, but there are two core problems with equating Scrum and "Agile."
First, there's a boatload of terrible Scrum out there that not only doesn't qualify as "Agile," it hardly qualifies as professional.

Second, there are many "Agile" approaches out there, and even though Scrum dialects are the most common, quite a lot of agile companies don't use even Scrum.

Inadequate processes, plans, documents, or contracts

This definition of "Agile" only rings true for illiterate people who can't make sense of the statement "While there is value in these things ..." - Agile approaches make use of these things. Any argument against "Agile" referencing to Agile's inability to structure themselves is either a strawman or ill-informed.

Lack of direction or tracking

Humorously, if that's the definition of "Agile," it's not a stretch to say that most large Enterprises are indeed extremely Agile.

On a more serious note: most professional Agile practitioners will emphasize the importance of clear goals, as well as frequently checking progress on these goals.

We're stuck with the same issue: if we use a strawman definition of "Agile," then everything sensible is a better choice - but none of these definitions hold true to the intent of the Manifesto.
The best definition I can come up with today is based on the very first line of the Manifesto itself: "We are discovering better ways of developing software by doing it and helping others do it." My definition, hence, is "The ability to do the right thing right, at the right time." And with that definition, everyone who proposes anything other than "Agile" - would have to argue why it would be preferable to do the wrong thing, do things wrong, or have poor timing: I find that case extremely hard to make.

Note that this definition of "Agile" says nothing about Frameworks, Methods, Roles or any other form of rules. It doesn't even say anything about values, principles or goals. It's so broad and encompassing that the dividing line is only, "Is that right?"

So - if nobody in their right mind would purposefully do something wrong, why is "Agile" even a thing?
And that leads us to ...

The actual choice

After this introduction, we realize that choice isn't Agile versus Waterfall. It's about answering the question how we deal with things going awry. But: why do things go wrong? Let's take Mark Twain's opinion here:

Mark Twain: What gets us into trouble is not what we don't know. It's what we know for sure that just ain’t so.

This provides us a powerful reframing of the entire debate: The "things we know for sure that just ain't so" are - unvalidated assumptions. With this framing, we can rephrase the question to:

How are we dealing with assumptions?

The fundamental choice is: do we proceed based on predetermined assumptions, or do we capture evidence to validate our assumptions? And at which point do we do this?

The framing thus stops being the unfortunate false dichotomy of "Agile vs. Waterfall," it turns into:

Which mechanisms do we employ when reality contradicts our assumptions?

A classical plan-driven approach is built on three core assumptions: we are doing the right thing, we know how to do it right, and we can determine the right timing.

A classical "Agile" approach is built on the core assumption that there is uncertainty: we don't know what the right thing is, we could be doing something wrong, and the world doesn't always follow our timing. The concept is the "VUCA World."

These approaches are both extremes, and in practice, there's always a form of compromise:

  • We can start with a hypothesis of what's the right thing, and when it turns out that it isn't, we need to change what we do.
  • We can define how we do things, and we have to change that plan when it no longer works.
  • We can schedule when we do what, and when we miss that schedule, we need to re-schedule.

Every project manager who learned their ropes will agree that even in a classical, plan-driven approach, this happens all the time. Likewise, every Agilist will concede that we need a product vision, some working agreements, and a bit of forecasting. We can also agree very easily that we need frequent inspection points and to check whether we're on track, and to change direction when that's not true. So - where does the disagreement come from?

Ultimately, we're talking about the probability that an assumption we made is wrong. How likely did we to bet on the wrong horse? How likely are we riding a dead horse? And how likely is the race not going as planned?

Statisticians understand that probabilities aren't 0 or 1, but somewhere in-between. Depending on how likely an event is, we should use practices appropriate to deal with the uncertainty. And depending on how far away an event is, the less likely we can predict it. Hence, the actual question becomes:

How do we deal with uncertainty?

The answer to this question is extremely contextual. We only know that rigid practices without flexibility will be more problematic when we ran into something that we didn't expect - and being so flexible that we can't make any form of prediction isn't sensible, either.

Summary

The question that we should be asking is how much it will cost us to learn that we are wrong? When the answer is tolerable, we are doing okay. Otherwise - we are gambling.

And in some cases, it could be the most wrong and costly thing to not have a detailed long-term plan. Even though that's not considered to be very "Agile."