Hacker News new | ask | show | jobs
by nine_k 509 days ago
Production of software is nearly 100% R&D. Making a million copies of a software product has a trivial cost. There are no assembly line workers in software (and the very word "assembly" means a different thing). A software engineer very often brings in revenue many times their salary.

Production of hardware is some R&D, and then actual manufacturing. Production of each physical item costs you. Production of every physical item has a chance to go wrong. Production of each physical item requires a number of humans (often a large number) to do repetitive, high-precision, high-skill work, as fast as practical. You can augment or replace some of them with robots but it also costs you, and you can't replace all the humans with satisfactory results.

So, with hardware, the cost of the workforce plays a major role, while with software it does much less. To produce physical things, you need a lot of people who are not well-off, and for whom factory work is an upgrade of their financial and social standing. A "developing country", with huge swaths of population leaving rural life for a better city life and factory work, is best in this regard. Ideally you sell your product to richer folks, maybe outside the country of production.

Of course there can be situations where the workers are highly paid, and produce very valuable things through their skilled work. Ford in 1950s famously paid the assembly line workers very well, so that they could buy the cars they produce, and valued their employment. But this does not always occur; people doing work that does not add a lot of resale value also want to live well, especially if the society does not want a flood of immigrants who are willing to work for much less. Check out how much the work of a plumber costs in Switzerland. So only high-precision, high-margin, low-volume manufacturing remains in Switzerland, such as precision optics, precision industrial and medical equipment, or premium mechanical Swiss watches. The US is in a somehow similar situation.

4 comments

I disagree very slightly. Mostly with this part:

>So, with hardware, the cost of the workforce plays a major role, while with software it does much less. To produce physical things, you need a lot of people who are not well-off, and for whom factory work is an upgrade of their financial and social standing. A "developing country", with huge swaths of population leaving rural life for a better city life and factory work, is best in this regard. Ideally you sell your product to richer folks, maybe outside the country of production.

You don't need a lot of people who are not well-off. You can automate the entire process. The problem with automation and labor saving technology is that it is capital intensive. The higher the capital investment per job (higher capital intensity), the bigger the chunk of money that flows to capital rather than labor.

This means that the cost of the workforce in a software company plays a bigger role than in a hardware company, where financing costs to pay for labor saving technology play a bigger role.

There are mining companies in Africa, who have nothing but an army of people equipped with shovels digging a small scale open pit mine. There is no way the labor cost here is the biggest constraint. An excavator and wheel loader could accomplish more with less people, but it would mean getting a USD loan to import foreign equipment and then selling for export to pay the foreign debts, rather than local production.

This all resonates very strongly with me. We have tons of automation - the proverbial "economies of scale", but we haven't managed to solve the last mile.

Auto assembly seems like a poster child. There's wild automation going on, but the typical plant still requires thousands of employees doing things by hand. Musk tried to automate a lot more of this away with newer/better robotics, but failed. (Tesla has still achieved a lot here, but it's been more towards creating designs that are more amenable to the current state of robotics).

IMO, this problem should be solvable now. I.E. we don't need "new physics" to reach another step-function in automation. We need more investment. We're still largely in the mindset of "special-purpose" automation.

> Production of software is nearly 100% R&D. Making a million copies of a software product has a trivial cost.

> Production of hardware is some R&D, and then actual manufacturing

Totally. And if you think deployment errors are bad, wait until you see how many production errors exist and how many items out of your line come out working and within spec

Indeed. You cannot release a patch for a mechanical part or a PCB.
For a PCB it’s called a rework, and it’s very common for first spins of boards to have to do one.

Also common is to patch around issues, when possible, in firmware. This is often lower cost/effort, but can’t fix everything.

There are similar kinds of fixes for purely mechanical parts. Depending on the part and problem, mechanical can be easier than a PCB rework (eg: modify a part in CAD and 3D print or get your local machine shop to do a run).

Or require a particular type of motor oil with a particular type of metal-based lubricant additive when you realise 100,000 cam shafts have shipped made of metal you’d assumed was to a higher spec but isn’t, just so the engine will make it through warranty period with insanely long service intervals.

I briefly looked at a couple used vehicles just outside of warranty and one within warranty that had literally had two oil changers in 100,000k, that’s 60,000 miles for the uninitiated.

You just release a new version. How many xbox 360s did they actually release? I think its close to a dozen iterations.

    > You cannot release a patch for a mechanical part
In NATO, this is frequent and normal. Many, many "recalls" are issued by military manfacturers, then local support staff spend X hours to replace the defective part. It is not so different from automobile recalls.
Correct. Also the economics of a mechanical patch are favourable for something in the M$ range with a fix costing in the 10k$ to 100k$ range
You can and people do.

It’s just a lot more expensive and labor intensive to apply.

Yeah I remember one of my friends working for a German auto company during the 2008 financial crisis and having insane stuff routinely happen like an auto manufacturer having to buy truckloads of sensors from a subcontractor that had nowhere to go as car manufacturing lines were stopped.

Failing to do so would have meant these manufacturers would go under, (along with their own subcontractors) and once demand shot back up, cars would be literally impossible to manufacture as key suppliers went out of business.