Respectable persons of this parish of Internet have been, shall we say, critical of Scrum and its ability to help makers (particularly software developers) to make things (particularly software). Ron Jeffries and GeePaw Hill have both deployed the bullshit word.
My own view, which I have described before, is that Scrum is a baseline product delivery process and a process improvement framework. Unfortunately, not many of us are trained or experienced in process improvement, not many of us know what we should be measuring to get a better process, so the process never improves.
At best, then, many Scrum organisations stick with the baseline process. That is still scadloads better than what many software organisations were trying before they had heard of Agile and Scrum, or after they had heard of it and before their CIO’s in-flight magazine made it acceptable. As a quick recap for those who were recruited into the post-Agile software industry, we used to spend two months working out all the tasks that would go into our six month project. Then we’d spend 15 months trying to implement them, then we’d all point at everybody else to explain why it hadn’t worked. Then, if there was any money left, we’d finally remember to ship some software to the customer.
Scrum has improved on that, by distributing both the shipping and the blaming over the whole life span of the project. This was sort of a necessary condition for software development to survive the dot-com crash, when “I have an idea” no longer instantly led to office space in SF, a foosball table, and thirty Herman Miller chairs. You have to deliver something of value early because you haven’t got enough money any more to put it off.
So when these respectable Internet parishioners say that Scrum is bad, this is how bad it has to be to reach that mark. Both of the authors I cite have been drinking from this trough way longer than I have, so have much more experience of the before times.
In this article I wanted to look at some of the things I, and software engineering peers around me, have done to make Scrum this bad. Because, as Ron says, Scrum is systemically bad, and we are part of the system.
Focus on Velocity
One thing it’s possible to measure is how much stuff gets shovelled in per iteration. This can be used to guide a couple of process-improvement ideas. First up is whether our ability to foresee problems arising in implementation is improving: this is “do estimates match actuals” but acknowledging that they never do, and the reason they never do is that we’re too optimistic all the time.
Second is whether the amount of stuff you’re shovelling is sustainable. This has to come second, because you have to have a stable idea of how much stuff there is in “some stuff” before you can determine whether it’s the same stuff you shovelled last time. If you occasionally shovel tons of stuff, then everybody goes off sick and you shovel no stuff, that’s not sustainable pace. If you occasionally shovel tons of stuff, then have to fix a load of production bugs and outages, and do a load of refactoring before you can shovel any more stuff, that’s not a sustainable pace, even if the amount of work in the shovelling phase and the fixing phase is equivalent.
This all makes working out what you can do, and what you can tell other people about what you can do, much easier. If you’re good at working out where the problems lie, and you’re good at knowing how many problems you can encounter and resolve per time, then it’s easier to make plans. Yes, we value responding to change over following a plan, but we also value planning our response to change, and we value communicating the impact of that response.
Even all of that relies on believing that a lot of things that could change, won’t change. Prediction and forecasting aren’t so chaotic that nobody should ever attempt them, but they certainly are sensitive enough to “all things being equal” that it’s important to bear in mind what all the things are and whether they have indeed remained equal.
And so it’s a terrible mistake to assume that this sprint’s velocity should be the same as, or (worse) greater than, the last one. You’re taking a descriptive statistic that should be handled with lots of caveats and using it as a goal. You end up doing the wrong thing.
Let’s say you decide you want 100 bushels of software over the next two weeks, because you shovelled 95 bushels of software last week. The easiest way to achieve that is to take things that you think add up to 100 bushels of software, and do less work so that they contain only, say, 78 bushels, and try to get those 78 bushels done in the time.
Which bits do you cut off? The buttons that the customer presses? No, they’ll notice that. The actions that happen when the customer presses the buttons? They’ll probably notice that, too. How about all the refinement and improvement that will make it possible to add another 100 bushels next time round? Nobody needs that to shovel this 100 bushels in!
But now, next time, shovelling software is harder, and you need to get 110 bushels in to “build on the momentum”. So more corners get cut, and more evenings get worked. And now it’s even harder to get the 120 bushels in that are needed the next time. Actual rate of software is going down, claimed rate of software is going up: soon everything will explode.
Separating “technical” and “non-technical”
Sometimes also “engineering” and “the business”. Particularly in a software company, this is weird, because “the business” is software engineering, but it’s a problem in other lines of business enabled by software too.
Often, domain experts in companies have quite a lot of technical knowledge and experience. In one fintech where I used to work, the “non-technical” people had a wealth (pun intended) of technical know-how when it came to financial planning and advice. That knowledge is as important to the purpose of making financial technology software as is software knowledge. Well, at least as important.
So why do so many teams delineate “engineering” and “the business”, or “techies” and “non-technical” people? In other words, why do software teams decide that only the software typists get a say in what software gets made using what practices? Why do people divide the world into pigs and chickens (though, in fairness to the Scrum folks, they abandoned that metaphor)?
I think the answer may be defensiveness. We’ve taken our ability to shovel 50 bushels of software and our commitment (it used to be an estimate, but now it’s a commitment) to shovel 120 bushels, and it’s evident that we can’t realistically do that. Why not? Oh it must be those pesky product owners, keep bringing their demands from the VP of marketing when all they’re going to do is say “build more”, “shovel more software”, and they aren’t actually doing any of it. If the customer rep didn’t keep repping the customer, we’d have time to actually do the software properly, and that would fix everything.
Note that while the Scrum guide no longer mentions chickens and pigs, it “still makes a distinction between members of the Scrum team and those individuals who are part of the process, but not responsible for delivering products”. This is an important distinction in places, but irrelevant and harmful in others. But it’s at its most harmful when it’s too narrowly drawn. When people who are part of the process are excluded from the delivering-products cabal. You still need to hear from them and use their expertise, even if you pretend that you don’t.
The related problem I have seen, and been part of, is the software-expertise people not even gaining passing knowledge of the problem domain. I’ve even seen it working on a developer tools team where the engineers building the tools didn’t have cause to use, or particularly even understand, the tool during our day-to-day work. All of those good technical practices, like automated acceptance tests, ubiquitous language, domain-driven design; they only mean shit if they’re helping to make the software compatible with the things people want the software for.
And that means a little bit of give and take on the old job boundaries. Let the rest of the company in on some info about how the software development is going, and learn a little about what the other folks you work with do. Be ready to adopt a more nuanced position on a new feature request than “Jen from sales with another crazy demand”.
Undriven process optimisation
As I said up top, the biggest problem Scrum often encounters in the wild is that it’s a process improvement framework run by people who don’t have any expertise at process improvement, but have read a book on Scrum (if that). This is where the agile consultant / retrospective facilitators usually come in: they do know something about process improvement, and even if they know nothing about your process it’s probably failing in similar enough ways to similar processes on similar teams that they can quickly identify the patterns and dysfunctions that apply in your context.
Without that guidance, or without that expertise on the team, retrospectives are the regular talking shop in which everybody knows that something is wrong, nobody knows what, and anybody who has a pet thing to try can propose that thing because there are no cogent arguments against giving it a go (nor are there any for it, except that we’ve got to try something!).
Thus we get resume-driven development: I bet that last sprint was bad because we aren’t functionally reactive enough, so we should spend the next sprint rewriting to this library I read about on medium. Or arbitrary process changes: this sprint we should add a column to the board, because we removed a column last time and look what happened. Or process gaming: I know we didn’t actually finish anything this month, but that’s because we didn’t deploy so if we call something “done” when it’s been typed into the IDE, we’ll achieve more next month. Or more pigs and chickens: the problem we had was that sales kept coming up with things customers needed, so we should stop the sales people talking either to the customers, to us, or to both.
Work out what it is you’re trying to do (not shovelling bushels of software, but the thing you’re trying to do for which software is potentially a solution) and measure that. Do you need to do more of that? Or less of it? Or the same amount for different people? Or for less money? What, about the way you’re shovelling software, could you change that would make that happen?
We’re back at connecting the technical and non-technical parts of the work, of course. To understand how the software work affects the non-software goals we need to understand both, and their interactions. Always have, always will.