There's also situations in the real world too, coin counting machines apparently also delay showing the results because people won't believe that dumping a jar full of coins into a machine and having an instant answer can be accurate.
People also do it too, locksmiths may umm and ah when called out to pick a lock because if they do it too quickly customers can feel aggrieved at a large invoice or even feel less safe in their homes if they've seen their front door picked in 10 seconds.
1 - https://www.theatlantic.com/technology/archive/2017/02/why-s...
Normal people have no reference point to judge how fast or slow software and hardware should be, other than through direct experience. By adding fake delays to avoid being honest and perhaps reassuring occasional surprised users, vendors just screw with the mental frameworks people build, at scale. It's a wasted opportunity, too, because if you're brave enough to be honest about execution time and weather the initial wave of distrust, then you may become a new reference point for your users, who will now view your competitors' software as bloated.
--
[0] - Though recently I brought this story up with a locksmith, and he said he thinks it's just stupid and there are hardly any situations in which it would even make sense.
1. Some disreputable locksmiths (word gets around) - probably very few
2. A locksmith will try the fast methods first, if they work he wins, if they don’t he didn’t waste too much time.
3. After doing a lock once he could immediately redo the same lock much faster and this may make it look like he could have done it very fast to begin
In some cases I could see that being the case: if the core task fundamentally requires significant computing resources and margins are too thin to through more resources at it while remaining competitive. That's the "every grocery store has one cashier with a long line" story (except for stores that change the story, e.g. with self checkout or a pricing strategy that frees up some margin to spend on more staff). I wonder how much of the slow performance of tax software is attributable to this, and how much is attributable simply to accumulated bloat or other things that might fall under the vague label "technical debt."
But it's also clear that there are technical decisions that can make the software slower for customers without making feature development easier or cheaper or faster.
(I am the person you're responding to) - I do think that very often it indeed falls into the vague label of "tech debt" but not all debt needs to be repaid.
For example - let's say I am writing a new Turbo Tax feature and I can do it quickly by leveraging 4 existing APIs. My feature is "slow" because each API does its own complex IO operations, many of which are redundant across the 4 calls.
Is that technical debt? I have the opportunity to create a new API that optimizes the IO for my use case, and it would be 4x faster. But, it may never be "worth it" to do that because my feature is seldomly used and is still fast enough for the user to not lose productivity. So I would say that's not really a tech debt, just a technical tradeoff that was made.
In general, I find that a lot of latency comes from (a) reuse of suboptimal pieces/frameworks that accelerate development and (b) not taking the time to optimize. There are times when these are tech debt, and there are times when these are fine.
Well, I'd still call that technical debt. All debt is a tradeoff, and sometimes debt is worth it (or, at least, is judged to be worth it at the time the debt is taken on). In this case, it might be deemed worth it, and reasonably so, especially if there simply aren't enough competitors that this would make a difference at the margin, or if it's simply not feasible for any competitor to implement that same feature in the same time frame with the same amount of developer resources and make it perform faster.
If you're FAANG there is some chance you're overpaying by not doing the RSU cost basis properly.
Now, you can definitely make the case that this is mostly Coinbase's fault for not sending over the actual purchase date, but TurboTax charged me a bunch of money to do that import, so I feel like they should have done some basic quality control.
I don't trust TurboTax to get my taxes right, because last year they got it very, very wrong.
What I'm objecting to is the idea that I'm not going to get my taxes wrong using TurboTax. They fucked up badly.
I'm certainly not arguing that the solution is to buy audit insurance from the same company that told me to overpay by several thousand dollars. Every penny I pay to Intuit is a moral failing on my part, and on the part of the legislature that succumbed to their lobbying.
What do you think the "real person" does? They put your data into TurboTax or equivalent.
A lot of people would live much easier lives if they believed this.
You can make a really half-assed filing and all they will ever do is send you an invoice for the difference.
In fact, you could simply not file your taxes ever and the IRS will eventually inform you of your actual tax burden, at which point you should probably pay accordingly. I've never gone beyond this point.
What you hear about and what they WILL get you for is deliberate fraud.
I gather from this comment that the software is only used by individuals filing their own returns, not by professional tax-preparers or accountants. The latter two certainly spend more than a "a few hours per year" in whatever product they use for preparation. When it's tax season and they have to get the work done by April 15, professional aren't going to tolerate a few more minutes per customer.
The main question is "does it matter?" The answer depends case by case.
For example, if I am the PM for TurboTax Premier desktop software, I know that it's latent - but it probably doesn't matter in a way that deserves prioritization. Sure, my clients would like for the forms to load faster, but they care infinitely more about the scope of tax situations we cover, and the correctness of calculation. A person uses my product a few hours per year, so even if there's an accumulated minute of form refresh latency, it's just not that impactful. What that means is - if I have an "extra developer", I am going to direct them towards product scope and correctness, not latency.
On the other hand, something like VSCODE (an example the article mentions), speed is part of the value proposition. As a developer, I need to quickly change text, quickly look up a reference, quickly switch files, etc. If things are sluggish, it isn't just frustrating but limits my ability to do development work. So a user of a slow IDE would be very motivated to seek alternatives, since they basically live in the IDE. As a PM, I would absolutely invest in performance if it was making my product actually less useful.
It's like anything else, investment has a tradeoff. Chances are, whatever car you drive could have a faster 0-60 speed and in isolation, that would be great. But are you willing to pay 10x for the car? Are you willing to give up the seats and the trunk to make it happen? So your car may not have the best 0-60, but if it's an affordable family minivan that's probably the right call.