The reverse is also true though -- hearing from folks that we need to measure nothing because "trust me". Both lead to poor outcomes, ime.
A good engineering team has measurements in place that are reasonable approximations, where it is reasonable to build them, but also treats them as prompts rather than absolutes. Asking "why is this metric out of band?" is infinitely more valuable than stating "this metric is out of band, we've failed".
Out of band means communicated on different channels than ordinary information. For example a metric communicated out of band would be told to you in person instead of showing up on your universal agent smith react widget dashboard. :-)
Two different jargon phrases overlapping here, as I’m sure you know. Yours is “out of communication band” whereas GP is “out of predictive band.” Yours is a bit more common to me, but I hear both.
A band, on a metric, indicates a minimum and maximum safe value. For instance, if you alarmed when your latency is below 50ms or above 150ms, that'd be your band. Being out of band is being out of that 'safe' window.
It's a totally valid and common jargon phrase from the web services world, apologies though, I assumed that it was wider jargon than it turns out it is...
There's no language rule that a phrase can have only one meaning. You seem to be taking this too precisely, as the earlier explanations of there being two interpretations are quite entirely reasonable.
It's also "you manage what you measure" - measure the wrong stuff ...
I agree with empathy and 'EQ', better still if we can measure the outcomes we really want (vs the easy ones). That requires knowledge and insight.
That’s also the kind of stuff I always hear from people that love to shoot from the hip and trust their gut feel. The results those people have is invariably inconsistent, there’s no way to scale any of their knowledge and honestly at the end of the day it’s just a person reacting to a situation.
You assume there is a way to scale certain kinds of knowledge, or that the current metrics-supremacy is it. I get it, humans are inconsistent and squishy, but at least we can be reasoned with. Stripping complex realities down to “metrics”, often without error bars even, gives fake authority to dangerous approximations. You can’t argue with numbers! They’re so precise and correct.
Discussions like these always feel too reductive to me. Both of you take extremist positions, when I’m sure you do both agree that the proper approach is obviously a middleground: Strive to gather objective numbers and measure things as much as possible to become able to make factual decisions, but don’t disregard human intuition just because it’s not measurable; after all, it’s a tool shaped by millions of years of evolution, culminating in creatures ruling the world due to their ability to assess social contexts correctly.
To put it differently—measure what you can; trust your feelings on the rest.
You can measure many, many things. Instead, you should measure things whenever they are good proxies for the outcomes you want to drive AND there is good cost benefit in measuring+analyzing them.
Of course I agree that the right thing is some sort of middle ground, of roughly what you lay out. But the current hegemonic position is that of metrics-supremacy, so I feel I have to argue most strongly against that to get us to that middle ground.
When you argue like that, that’s what’s generally referred to as “pushing an agenda” and people sour to it quickly. You’re better off sticking to your actual argument than embellishing things.
I'm not embellishing anything, I'm making my "actual" arguments. I've never argued against seeking a sensible middle-ground. I've said elsewhere that metrics have their place.
The world isn't black or white, zero or one. The main problem with metrics-centric management is that its proponents try to reduce everything to metrics.
I was simply pointing out that many situations can't be solved through metrics. In my experience, the best managers try to solve most problems without metrics, but always insist on them in places where they are valuable.
In other words, they use metrics as a tool and make metrics work for the team instead of the team work for the metrics.
Given a choice between two metrics, managers will usually choose the most gameable, indirect, activity-oriented metric, even if it is far less useful for driving results.
And I can’t blame them. That’s the game. They have to make a number go up, their boss also has numbers that need to go up, and so on. In the worst cases, everyone is complicit in creating a forest of bullshit numbers that went up even while the company tanked.
What alternative do you see? At the end of the day it is philosophically impossible to manage situations that cannot be detected. It is like the OODA loop to catch a ghost; the ghost can't be observed because it doesn't exist and the ODA part becomes a questionable exercise.
Managers who react to known signals in predictable ways are strategically better than managers who just make stuff up on the day in unpredictable ways.
To use a famous, well-known example relevant to the HN user base: Software productivity can't be measured with any simple metrics - essentially every attempt at coming up with software output metrics can be trivially gamed, yet that hasn't stopped people from trying mightily (I'm probably showing my age in recalling how much effort was put into "function points" from a couple decades ago).
At the same time, nearly all of us can identify developers who are highly, highly productive, and others who are not. We may not have great metrics for software productivity, but I can guarantee that Fabrice Bellard is about 100 times more productive than I am.
Metrics are inherently an oversimplification in attempt to get complex phenomena down to a strictly ordered value, but the real world doesn't work like that. I think your mistake is thinking that just because things can't be reduced so a small set of numbers would also mean they can't be observed, and that isn't true.
You also have a problem that a team can benefit from people who are less productive as coders, but more productive at paperwork, organizing, mentoring, meetings, etc.
I’ve seen several teams fire less productive coders and then the team fell apart into individually good coders.
That teams require a blend of capabilities and personalities is anathema to the “everyone a widget” mentality of most corporations — but remains persistently true.
As a manager, there is diminishing returns to measurements. At a certain point, you have to trust the team to do the right thing without having full documentation, specification, and provable outcomes.
Many of the most important decisions do not have quantifiable outcomes. How many incidents did this architecture or refactor prevent? What cultural effect would that person we didn’t hire have had over the next year?
At least in early stage or growing startups, I think if every one of your actions is measurable and fully specced, you’re in trouble.
> At a certain point, you have to trust the team to do the right thing...
Yeah, but that is explicitly not managing them. That is the standard "I can't measure this situation, therefore I will not attempt to manage it" response of a good manager.
You are not perfect though. There will always be factors overlooked and simply acknowledging that before it becomes a problem is something most management has failed to grasp which is what motivates this sentiment. I can bet now often than not on this thread alone now people have felt the negative effects of every action being quantified for the sake of 'profitability' by someone who lacks a firm grasp on the field being measured, which leads to high performing employees being let go.
I don’t really know what that means. Management is not about control, it’s about outcomes.
I am responsible for the outcomes of my teams, full stop. I am responsible for the growth and development of the engineers who report to me.
If that means sometimes standing back and letting the cook, then that’s the right thing to do. I’m not here to involve myself because otherwise people will ask “well what are you doing??”
(The answer to that is that the more independent teams can run at a high velocity of quality output, the fewer managers your company needs. That is good!)
> Management is not about control, it’s about outcomes.
Management is about controlling the outcomes. Nobody could claim to be managing a situation well if they are responsible but exercising literally no influence over the process. If the manager could be replaced with a garden gnome and it doesn't change the outcome then the situation is effectively unmanaged.
We seem to be in furious agreement so I don't know why my perspective is eluding you. You're looking at a situation that might have an outcome that is unacceptable. You've identified that there is no measurable way to assess progress or quality. You've concluded that there is nothing you can say, do or observe that will help. The situation is then left unmanaged because there is nothing to measure and you're waiting for something you can measure before you act. This is the basics of management, what you can measure you can manage what you can't you can't.
It just happens that in the small software is unmanageable, as quickly becomes clear if you have management experience of a non-software process and compare it to managing a software team.
You're claiming that it's "philosophically impossible" (whatever that's supposed to mean) to do X, as if philosophy is some immutable and fully-understood constant like a gravitational constant. This is what reading too many management books and theories does to people - management theory isn't some sort of physical constant which is unquestionable or even testable
Detected = measured. They are equivalent actions. You can even trivially metricise such a situation - plot it on a graph, x axis time & y axis a plot of "did we detect it?".
Are the companies you most admire run that way? Is that how the technologies that you love were invented? Is that how the individual technologists you look up to operated? Is that how the most impactful, world-altering inventions were created?
For me, the answer is no. Perhaps for you the answer is yes.
You are not thinking at all carefully about what it is to measure something and how much meaning you get from that. Metrics need to convey a lot of useful meaning, and measure something that is worth measuring, otherwise they are worse than useless. I’ve seen otherwise-intelligent people led totally astray because of this totally blinkered fetishisation of metrics over everything else. You cannot meaningfully reduce even a small complex system to a handful of graphs. There is a place for metrics, but responsible use demands humility, not supremacy.
Our ancestors made do for thousands of years, hell they coordinated the logistics of wars with far greater complexity than even the largest of companies.
It’s one thing to suffer from a blinkered, reductive view of the world. It’s another thing to be proud of it.