Efficiency is the enemy? (2021)
fs.blog
fs.blog
Always have some free disk space. Do not run your web servers at 100% CPU utilization. Overprovision your clusters. Add redundant network links. These ideas look completely obvious, don't they?
Now apply them at the org level. Always have someone with spare cycles. Leave some time at the end of the sprint when all planned tickets are done. Allocate time during the day to sit and think, or to experiment, or to take a walk. Let the person oncall to mostly sit around and watch the system operate instead of churning out something. Now this sounds a little bit crazy, doesn't it. So much efficiency is left on the table!
(Also, certain Nassim Taleb used to be popular around here; he wrote a whole book, "Antifragile", based on these ideas.)
I think he'd probably be against "Max efficiency" thinking, but "Max effectiveness" thinking is still not the same as "Minimum major trouble". He's a bit too "Move fast and break things", which is kind of like max efficiency, if you define efficiency as "How much business growth can I pack in ten years".
I've always thought that things should be planned with "Time firebreaks", that are absolutely excluded from all ahead of time planning whatsoever, with the intent of cleaning up any unexpected trouble you run into. Not quite a break time, those are planned ahead and known to be necessary.
Jacques Ellul, for example, posited that our "technical system" aims at efficiency to the point of engulfing everything, in a detrimental way because (composed with physic laws) it leads to betting on economy of scale (gigantism, centralization, specialization...).
Stephen Jay Gould "criticized the belief that evolution comes up with the best, or only, solution to a biological problem", and that organisms are always working towards fitness.
Leaving a bit in the tank every day increases ability to respond to sudden emergencies more intelligently and with less burnout afterwards.
A much better example of the benefits of slack would be a CEO who delegates the day to day work to subordinates (which includes "making decisions", the thing the article says Tony's main job is), to give himself slack--more time to think about big picture things and consider ways to change the organization to make it better.
In the case of Gloria, she should have some amount of slack to figure out how to do the big picture items of her job better, which include how better to help Tony.
Where things get tricky is when looking at the whole organization there's been a long tendency to see "slack" as "fat" that's to be trimmed. So, if a organization has two Glorias who both have some slack time, the argument is the organization would be more efficient and profitable having one Gloria with less or no slack time.
This has worked for corporate profits, but it leads to the humans in these positions unable to have the slack time needed to improve those positions.
Sometimes, this is fine and you can build a very profitable business off of it. For example, McDonald's isn't looking for their frontline staff to do any thinking.
In general, if a position can exist and doesn't need "slack" time, it is probably a position that should be automated as soon as it's cost effective.
If you really need a human to do it though, you should stop trying to treat it like machine and realize the importance of slack time and build it in to the schedule.
They tend to punish it. I say that I learned from fast food: how to mop a floor, how to remove burnt coffee, and the value of an education/career.
I think because they work with a lot of teenagers, they tend to get away with belittling people and getting because you’re two back-talks from fired, and you’re not sophisticated enough yet for the subtle burn they can’t write you up for.
“Give a man a little power” was never more clear to me. They have their little empire of dirt and they are going to defend it.
The biggest asshole had a brother who was a pro baseball player. As if what your brother does makes you important (my own brother hated that idea). Congratulations, you sell burgers all day. Your parents must be proud.
The best of them had a mild case of impostor syndrome he tried to cover. Told too many dad jokes, but still top 3 for managers in a clutch situation. I hope he did something with himself besides ace school field trips dropping in on us unannounced.
You could imagine a scenario where everyone at the company is taking a break when X has their issue, and that's the best case scenario. Instead of everyone off doing their own thing, they're slacking off but ready to help. The combined context switching in an extreme case like this could be hours or days.
Just like running 100% lean (just in time efficiency) reduces overall resilience when snags occur. Having SOME slack (available capacity) allows for agility when things go awry.
https://blog.danslimmon.com/2016/08/26/the-most-important-th...
That's not slack. As you just pointed out, that's his regular job. Slack is unscheduled time, not time that's devoted to performing your duties. The point of slack, anywhere, is that it lets you respond to situations that arise, by converting the unscheduled time into time in which you respond. If you're fully scheduled, that's difficult to do.
Time that the CEO can spend thinking about big picture things and considering ways to change the organization to make it better is "unscheduled time". That's the point.
We can easily state this another way. Slack is unused capacity to do work. The CEO planning his strategy is work, the opposite of slack.
The CEOs entire enterprise is due to others. If anything explicitly highlighting that undermines the myth they’re the experts.
Both of your models rely on the CEO delegating to focus on the right state change for the business, so I’m not sure why you create two scenarios with squishy ideas of a CEO being efficient or slacking. Some CEOs have casual demeanors, others addiction to Adderall energy; neither is slacking if that’s just the vibe they need to “make it better.”
You seem to be circling the core meme it’s the CEOs job to improve the company and wrapping it up in subjective modals for doing so. Not sure how this refutes the article.
Most jobs don't set salary based on the necessity of having someone in that role, it's usually based on how difficult it is to find a capable person in that role.
If there are 100k people capable of doing the job of the secretary, and 5k people capable of doing the job of the CEO, then the CEO will be able to negotiate a higher pay rate.
It’s not the only viable one to the species.
Given automation and first hand experience building analytics systems for VCs and investors; they’re not that much more capable. They’re humans at the same edge of knowledge as any grad of a decent university, of which there are millions.
Teaching “stay in your lane” tacitly props up apathy and disbelief in the proles.
Slack is the ur-antifragility.
(efficiency pursued for itself is AN enemy)
Slack is time & permission to consider, reflecr, grow, enables finding new ways to thrive.
If risks werent treated as an externality this failure mode would be taken care of. Of course some unanticipated failure modes would always come and surprise us. I would be one happy person if the failures we see around us were all of this 'inconceivable' variety. Very few are, except when one buries one head in the sand and refuses to consider the fairly obvious risks.
For example, the financial entity that makes tons of money speculating in a worthless asset, then goes belly up when its worthlessness is exposed (I won't name the asset(s) to avoid a flame war).
You can't "price that in." You just avoid it.
Avoiding it falls in the same spectrum. It is equivalent to assigning an infinite cost to the risk .
I like this piece. It speaks to an efficiency measure very often overlooked!
The way I see this is everything comes with costs and risks. Good systems manage those.
But they are not entirely eliminated.
And that's the game! Running full capacity is more expensive than some reduced capacity, and what that number is depends on what an organization has incoming.
Low latency responses to high priority events very seriously reduces overall costs. Having some degree of capacity to deal with it makes a ton of sense.
Invariably, someone else sees that as an unused resource.
The argument will be made that Gloria could actually be doing a lot and simply drop it when Tony ends up needing stuff.
And while true, task switches are not free, nor is basic human readiness. For Tony, he values knowing Gloria has command of the details and has good processes she has likely used idle time to conceive.
For Gloria, jumping into action is expected and normal. She is not haggard into a crappy state every day.
Efficiency has different measures.
I have seen similar in manufacturing where any capacity left unused seems like money left on the table. But the truth is people are often employing ideas to make their jobs easier, bring quality up, and the like.
Sometimes parts need rework. Stuff happens. Doing that may well be transparent to others given there is some capacity to get the rework done.
1) Efficiency hates redundancy
2) Redundancy is, like, the easiest, most zero-thought way of adding robustness
I've been a little suspicious of efficiency.
And we see questionable efficiency optimizations all over -- here in an individual business, in the greater economy (Let's over-specialize our economies, it is more efficient. Let's JIT manufacture everything, buffers cost warehouse space).
Just add redundancy. It'll be twice as expensive but it won't explode if something goes wrong.
For example, maybe you can use something like the Reed–Solomon error correction, which allows you to balance between no redundancy and full duplication, both in cost and benefit.
https://en.wikipedia.org/wiki/Reed%E2%80%93Solomon_error_cor...
Optimizing for latency will often improve system throughput, and maximum throughput almost never shows 100% utilization.
1. You have a lot of wasted work. You're making 100 widgets/hour that the next station can only consume at 20/hour. Inventory piles up and what happens when that work product is no longer needed or expires (if it can expire)? Now you've spent more (on the inputs to production) and have spent money you can't recover in full.
2. You cut to the quick. You look at that previous situation and say, "Let's sell 4 of the 5 machines, we only need 20/hour so why are able to make 100?". Turns out that the extra capacity helped you deal with maintenance downtime, tool changing, quality control issues (low quality parts, which should itself be addressed anyways), or meet customer needs for spare parts (infrequent, but large orders like at one former employer of mine).
(2) is what happens when a lot of people learn of JIT but don't comprehend JIT (or Lean). They don't understand, by analogy, that cutting your arm off to lose a few pounds is probably not the best idea out there.
My prior employer learned this, they cut out the travel and finance teams (about 8-10 people in total) and moved all that load onto the engineers who traveled, individually, infrequently but as a group very often (1500 of us). Problem: They did it infrequently, with a crappy web app. You went on a trip that lasted 1 day? It took you 4-8 hours in advance to set up everything, and 4-8 hours after to put in your vouchers. So a single day trip could take 3 days of total time. And for the first time someone used the system, or if they had anything unusual in their voucher, it could take days (worst I saw was a week of 3-5 hours a day spent on their voucher dealing with the back and forth on getting it approved and amending the voucher). They brought back the "idle" travel and finance teams, and, huh, suddenly we weren't wasting so much time on it. They were experts in the system and could get those complex vouchers done in a couple hours, and the simple ones in minutes.
It seems to me like they misunderstood the value stream. The specialized added value by reducing the average administrative overhead of the engineers, which enabled the engineers to focus more on where they add the most value to the system. Not unlike the secretary and CEO in the article.
Because of their specialization they were fast with broken systems (they used the same crappy web apps as us, but they understood them). Because they were not engineers and they got the work done fast, there was no perceived value in their role (classism and other biases at play here), it was thought (bad assumption) that the engineers could do it just as well themselves. And since the engineers had more than enough slack anyways, they could lose 15-30 minutes before and after a trip to dealing with the paperwork.
And if that assumption had been correct, maybe it would have been the right call. But the assumption was incorrect, and it was a very stupid call. We didn't charge the customers for that overhead, so any time the engineer spent on it (hours, sometimes days) was revenue lost. The dedicated team cost much less than the lost revenue so it was overall a net loss.
It really does the reader a disservice by not giving them the temporal context in which the words were written.
[1] https://www.amazon.com/dp/0199729883
[2] https://www.lesswrong.com/posts/zpA2Tnp2k38qSmr8J/how-to-ide...
> A world without slack is not a place one wants to be. Mazes systematically erase all slack. Slack is evidence of not being fully committed, and given that everyone’s skills are equal and competition is perfect, holding anything back means losing even if undetected. [1]
And an article by Zvi on the specific concept [2].
My take, and my experience, says that a lot of what makes a moral maze is an organization where professional managers engage in a large-scale effort to ensure that blame or negative consequences flow past them. So you advance by building a maze around yourself so that negative results miss you and you grab credit for positive results that have to flow through you.
When people "on the line" do good work, you can leech off of that. And as long as your organization does good work you can do fine. But if your organization becomes marginal or if you're trying to outcompete other middle managers, you have to find a way to pass that downstream. As a generic "manager" you can't add value on the line, but it's very easy to blame lack of results on lack of efficiency (too much slack) so that's what gets sacrificed.
[1] https://thezvi.wordpress.com/2020/01/12/how-to-identify-an-i...
Efficiency is only useful until it imposes on some degree of slack, or spare resource, which allows you to respond to aberrations which (inherently fragile) efficient systems can't accommodate very well (Although you could argue that an efficient system would accommodate systemic aberrations well, but then I wonder if it would be a semantic argument).
Sort of like with finances. You could optimize your spending and investing as though economics are totally determinate, but they aren't, so you keep some cash as slack to accommodate cases where the shit hits the fan.
When working on a project, you could schedule your time down to the wire and work as hard/smart as possible to hit the deadline perfectly, but it's not going to work out well every time (or often at all in software). So you figure alright, it seems like something that should take 3 days, I'll bank on 5. If you come in over or under, it works out well for everyone. There is not point in trying to max out on efficiency in these cases.
In determinate environments where things really do work exactly as you'd expect, well, it's a different story. But human beings aren't that.
Pro athletes and ultra-high performers use these types of metrics to maximize their "potential," but it seems to come at the cost of burnout, injury, and things like mental health and family issues.
I get a pretty significant amount of stuff done, and pretty well, but I could care less about measuring it.
I don't compete with anyone; especially with myself.
When crunch time comes, I roll up my sleeves, and get downright obsessive, but I also do things like take "side trip projects" (for example, breaking out UI widgets into open-source packages).
I get done what needs doing. I learn a lot, and I keep myself relatively healthy.
Some of them acts as very useful "information providers", they know who knows and "productive" members ask questions to them, and at best some are able to provide useful information even before being asked to do so.
Without them the amount of resources wasted by information vectors (service notes, meetings, mail, chat...) grows proportionally to the size of the organization.
One would think that a life-long fascination of zen buddhism would have taught me that already but I guess not.
I prefer talking directly about inputs/costs and outputs/desired effects. This also makes it more explicit who pays the costs and who gets the benefits.
That's the core. Once understood that the rest start to be obvious.
“You’re efficient when you do something with minimum waste. And you’re effective when you’re doing the right something.”
This is what's called productivity. Efficiency is how much the productivity align to a certain goal. Depending on which goal you measure the productivity against the efficiency will be different.
In the case of Tony "Every minute of his time goes on the most important part of his work—making decisions—and not on dealing with trivial inconveniences like waiting in line at the post office."
does he take the -right- decisions for the company benefit? Could he think up good strategies at the waiting line? or get inspiration from the surrounding? or just rest at the waiting line to perform better in action?
To make it more trivial: Imagine a worker that's really good at making shoes. This worker can output 60shoes a hour (with good quality). But now this worker actually works at company making computers. The computer company does not really now what to do with all these shoes. The worker is really good at making shoes and have high productivity compared to his fellow friends at a shoe company. But it does not help the computer company to sell or produce more computers. This shoe maker would probably be a better fit in a shoe company. (But who knows, maybe the shoes are so good that the fellow computer makers need less time at the doctor by wearing them. And them seeing the shoe maker creating shoes with such determination and productivity, that they get inspired to make computers with same determination and productivity.) Only the results can tell.
So what is the goal? That is probably the most difficult question. As to be able to measure efficiency the goal needs to be measurable. And probably there is not only one goal that a company strive for but many. And these goals can compete with each other. These goals can also be divided into smaller goals, for different parts of the organization.
Efficiency isn't so easy to measure but productivity is. Therefore most tend to stick with productivity as a tool to improve.
There's often a mix up with productivy and efficiency which makes the messages confusing.
A good read: M Goldratt, E. and Cox, J., 1993. The goal - A process of ongoing improvement.
https://www.amazon.com/Slack-Getting-Burnout-Busywork-Effici...
Once you have a process that works, then you can focus on driving more efficiency out of it.
AKA watch out for premature optimization.
https://www.tbray.org/ongoing/When/202x/2020/07/05/Too-Effic...
on this topic. Tim Bray's writing is always worth reading, but I think he's really valuable on this specifically.