Think there’s a book called “measure what matters” and the idea is what we measure shapes companies and behaviour. So I’d be very careful about implementing anything like this in my org.
This is tackling one aspect of this - highlighting the cost to the engineer.
It adds things like management of guardrails and policies that can be maintained by a finops role, or engineering manager. PRs can block or notify if thresholds are going to be breached by a specific PR.
As an electrical engineer, we scrounged for every cent on devices when they were being released. A TVS diode can be replaced with a capacitor? That's a cent. A cheaper processor? Several tens of cents. Over ten million devices plus the legacy of your design into the next generations, that's a lot of money. Someone else negotiates the price of the TVS diode and the processor, but that doesn't mean you should be isolated from the cost. A thing that does X and is more expensive is worse than a thing that does X cheaply.
Software engineers spend fractional cents on requests executed thousands of times per second. That's the same scale as the electrical engineering example when you do this.
As far as I can tell, most good software engineers are very cost aware. That doesn't mean they don't do stuff. They just understand the cost of that stuff. This does seem to be a big divide between senior and junior engineers, as well.
Because it's not their job and therefore they don't have the information necessary to know what is or isn't expensive.
Their job is to come up with the best design using common sense when it comes to cost.
Cost of a part isn't the only thing that matters either. Ok, we can save 1 cent if we swap this part out. But now we have to purchase lot sizes that are 10x bigger. How does that impact production? How does that affect operational budgets? Etc.
The engineers engineer and the bean counters count beans. They meet and find a happy place.
Why would you even want that extra workload? The fact that you think it's so easy just goes to show why actual accountants do the accounting, managers do the managing, procurement does the procuring, and operations does the operating.
"I know calculus, accounting is easy!" Sure, if you throw out all of the variables other than unit cost.
Yes, you shouldn't care about lot sizes and cost breaks (unless they are significant deciding factors), but knowing that one operation is going to cost 2x another, largely equivalent operation is relvant/important.
Anyone can make a bridge that stands. It takes an engineer to make one that barely stands. The same largely goes for computer systems.
Those same developers don't get to complain on the rainy day, when someone who knows how to save the company millions of dollars a year gets a better deal.
This is why bridges are not just huge chunks of titanium. Those would work very well at handling 50 tons at year 70, but would be shitty bridges because they are too expensive to build.
Could it possibly be because it is cheaper to insure, repair, and work with those designs?
Risk is expensive. That's why you reduce it. If it weren't expensive, you would take it on.
That’s not necessarily the risk being avoided. The customer may be better off, but project management’s risk vs reward looks very different than the customer’s risks and rewards. Get enough stakeholders together and …
So no, people really do regularly reject the cheapest design that fulfills the spec.
That being said, you should be able to have enough awareness to avoid anything particularly egregious - eg. avoid things that you would never want in a final product anyway in the first place. I guess we probably agree on that when you say standard parts etc.
It is your job, when you make something, to understand how it works. How much it costs to operate is an integral part of how it works.
> Ok, we can save 1 cent if we swap this part out. But now we have to purchase lot sizes that are 10x bigger. How does that impact production? How does that affect operational budgets? Etc.
A 10x larger lot size makes each part per unit cheaper and it simplifies production to eliminate items from your BoM. Understanding that sort of thing in an order-of-magnitude sense is also part of your job.
By the way, you don't need to know this stuff precisely to make good decisions. You need to know it roughly. You still need to know it, though.
> The engineers engineer and the bean counters count beans. They meet and find a happy place.
This is eventually taken to the extreme and you occasionally see posts here titled something like "how we saved millions by doing X", where the thing they were originally doing was extremely wasteful.
For the last decade, we've also been in a "free money" mode. Where companies were happy to spend money as long as it led to growth. Optimizing for cost wasn't a priority.
That has leaked into electrical engineering as well. There were a few products that shipped full Raspberry pis in an enclosure in the name of velocity. The savings possible there were probably in the dollars, not cents, not to mention the supply chain issues. And yet companies did it in the name of velocity.
Sometimes it's better to have a product that you can sell, than waiting to perfect the product to sell.
Let's leave aside the RPi availability issues, you could do this with any other discovery board.
Say you want to sell your first 1000 of products x, y, and z to see how the product market fit is. You might find that x and y don't sell at all, but z was sold out in the blink of an eye.
Now you can design a custom PCB for z, knowing that it will probably sell well. And until that is finished you continue selling the off the shelf version. Some profit is better than no profit, and consistency is good for your brand.
Software engineers, especially at startups, are building proofs of concept that they need to iterate on very quickly. Spending significant thought on reducing opex is an absolutely terrible call in the early stages of dev.
If you have a profitable product, or one that would be if you can squeeze costs, then you play the optimization game.
The difference between electrical engineering and software engineering is that software can be updated after it’s deployed. The ability to get all of those gains later changes the strategy completely
I fixed that one for you. Software engineers not at startups are usually not building proofs-of-concept. They are usually building production systems. And at the early stages of development of production systems, it can save a lot of headache down the road to think ahead. Yes, you will be slower in the moment, but the number of times I have heard a startup say "we did great with our first version, but then realized we needed to completely rewrite the thing to scale" (when the rewrite takes literally a week longer than the first version) has been too high.
This is completely false. Most startups go through massive pivots. I’ve been through 6 of them (2 successful exits) and literally everyone made drastic changes to every code base to target completely different markets and features.
I’m telling you from experience, if you can’t recognize providing business value as an engineer recognizing opportunity cost, you will stagnate as a junior engineer or even destroy the company as a senior one.
This attitude is one vendors have leaned into way too hard, and it's why there are so many consumer products out there which shipped with half-baked, buggy firmware.
>which shipped with half-baked, buggy firmware.
Because consumers pay for this shit. It’s the world we live in. The people working on the bug free version ship a year later after the buggy competitor ate the whole market. They’ve also taken the recognized revenue to take a higher valuation and hire more engineers to rewrite their buggy system than won the market.
Bugs don’t matter unless it’s a competitive market where they can change a sale. A startup operating in that kind of market is already fucked.
Personally, baking this kind of spreadsheet work into a pipeline was a moderate pain in the butt and hazardous to fully automate.
The dev doesn't worry about it unless it is obviously crazy, I.e. above a team budget. If so they sanity check the code.
A company employees engineers in the hopes of reaping value significantly higher than their pay in the future. Analysis takes time, time requires money - reviews require multiple people's time. If you are looking at 1k per month in expenses - but spend 1 month debating methods of saving ~500 bucks per mont at the cost of ~1 engineer. You can easily be looking at a 20+ month time to return on investment. Computers generally halve in cost every 2 years - so you're time to reach a 2x ROI can easily be 4-5 years. Most businesses need a return on capital of ~10% to justify investors not simply placing their funds in the stock market.