Hacker News Reader

Comment by osigurdson | Hacker News Reader

>> 5 people on a system that has about 40 users

Gosh, that is expensive software!

That's weird way to look at it. It's about value, not cost. For all we know that 40 user system manages the entire logistics of a national shipping company and saves them 50x the ops cost. Still think its expensive? Or maybe it's a corporate risk register with significant potential regulatory and reputational harm attached to each risk. Still costly?

I didn’t intend to suggest that the software does not have a positive ROI.

Some applications (control systems, etc.) have no users the traditional sense.

not so much. at the last investment bank i worked at we had about 10 developers writing code to support about 10 traders, perhaps a few more. i don't think this is unusual. this is complex stuff that is changing all the time due to regulatory and business issues.

Oh that's nothing. I worked on a project at a fortune 500 footwear company with dozens of developers, FTEs and contract, that worked over a year on just an update to a software package that was used by maybe 5 people at the company. Granted, these 5 people were the folks actually designing and figuring out how to manufacture their shoes, but still. At least 20% of the contractors were just warming their chairs, and more than one were what I call net negative producers: anything they wrote had to be corrected by someone else before it could be called done.

Seriously. It’s hard to imagine a group of 40 people that would need a sufficient velocity of changes/new features to warrant 5 full time engineers.

Massive corporate HR Dept with custom rolled software. Bank team with software for compliance. Endless possibilities...

From my experience, this is pretty common in AI

Internal tools used by a dozen or a couple dozen AI engineers/Data Scientists, the models they output used by millions. It's totally worth keeping 5-10 person engineering team to create better and stronger tools that make upgrading and deploying new models easier and faster

> totally worth keeping 5-10 person engineering team

I've worked in AI/ML for a long time, and I can tell you that in all but the rarest of exceptions nobody has sat down and done the math to really determine if that value is there. In my experience, most cases it's not.

I've seen teams like you describe and then asked them "what's the difference in value between what our model does today and the theoretical optimum"? The answer, which nobody liked, was fractions of a penny per user. That means if the team achieved perfection it wouldn't really matter in practice.

We're going to see a lot of the AI/ML teams disappear as companies are forced to focus on determining the real value add being provided.

Imagine each of those people is an employee at the same company, and they each command a salary of say 100k. How much more productive do you need to make them for it to become worth it? If you 3x their productivity, they are now doing the job of 3 people, you just saved 200k _per year_.

To tie that back to your comment -- imagine its for a team of 40 users but the company would ideally like them to be a team of 400 users. Yet because of the software being created, those 40 people are doing the work of 400.

You can extend that line of thinking pretty far -- and going down that line of thought is when I realized how valuable programming is. When you build the right things you are literally creating value with everything you ship, and it adds up over time.

>> when I realized how valuable programming is

Definitely. A cheap computer can outperform a billion humans at certain types of tasks. One must merely find the right tasks for the computer in order to exact unfathomable value. AI is going to expand the problem domains that computers can compete in by several orders of magnitude. The era of SPUs (Single Person Unicorn) companies is near.

Any trading application, where markets are constantly changing, new financial products included, new models developed etc.

Every company needs internal software of some description, and sometimes you're doing unique stuff that can't simply be purchased off the shelf.

I mean the extreme case is a developer writing scripts they only use themselves, which would be a 1:1 ratio.

Surely the extreme case is whole teams working on software only one person uses

I mean you're right but also; what does a manager do if not producing "software" to a small group of individuals. Or a designer for that matter.

Anyone who produces direct output to a small group of people is expensive in a similar way.

There's a really interesting phenomenon at play here whereby osigurdson said that something was expensive (and it's probably about $1M/year/users) and everyone has replied as though he/she said that it was too expensive.

You are probably right, but not nearly as expensive as the 40 users (assuming employees). So if you can double their productivity it is worth it.

It could well be this software is what makes it 40 users and not 4000.

Or a really really value-add internal tool!

You can't tell that from the data provided.

Reply on news.ycombinator.com