Hacker News Reader

Comment by justahuman74 | Hacker News Reader

What are some examples of things being built that are not adding anything to the company?

I worked there a few years. The thing is, some systems are so complex it might take 6-9 months to even understand the value add within the org. At which point it comes to light that there is very little value add. And all the meetings you were kept from attending were political gatekeeping.

An example would be a recommendation system who’s recommendations are ignored by downstream systems. And yet, 30 people in an org work on that engine.

Did people working on that engine know that their work is being ignored?

Yes…but the thing is downstream systems are dynamic and real time, so “could” use the recommendations, but never actually do. Hypothetically…

This type of thing goes on all over the company. Another example, a complex “data provider platform”. Which is actually just spinning up spark clusters, joining data, and making it available to downstream consumers. There’s a lot of bells and whistles, notifications built on top. Strip away the details, it’s just doing a join. Ten engineers, two senior engineers making 500k, long hours.

That’s Amazon.

I encountered the same sort of problems at amazon, I felt like it was because leadership didn't incentivize those optimizations to be made across the org. We'd identify some team that was the only one that could help us with something and nobody would lift a finger for us. Teams who were on the hook would ship us incomplete and untested APIs claiming they were done in order to work on other stuff. Managers didn't want to bring it up with other teams because of political ramifications.

That’s insane, how do the people in the org justify their work for promotion? There isn’t any need to map it back to observed top line metric shifts even for a 30 person effort?

“Person xyz built a complex model deployment system, capable of updating the recommendation engine in real time, with automated rollbacks and A/B testing. He leveraged 4 junior engineers and collaborated with multiple data science teams”

Indeed, this is why corporate KPIs exist.

Requiring everything to move KPIs is psychotic, but if your effort should have moved a KPI, and didn't, then you'd better have a damn good story about how it delivered value elsewhere instead.

I think you are describing all big tech companies. But what's the solution? I think it's just a result of having a large org. 20% of the org delivers 80% of the value. But when risk appetite is high and the company can afford to throw money at experiments, it makes sense to still pay for the rest of the 80%. When times are tough, it doesn't.

Definitely it makes sense to take bets on new tech. But a lot of higher up managers are not running these projects in good faith. At some level they know the projects are futile, and many around them are playing a game acting otherwise

How do you know that? I don't disagree that many projects are futile. But it is usually only apparent in hindsight, and humans are very good at justifying their own jobs and by extension, their team's jobs.

> But what's the solution? I think it's just a result of having a large org.

Right. Maybe Elon's approach is correct, and all these layoffs are a chance for companies to trim the fat and focus on what matters.

We're only now realizing the bubble the tech industry was in, and that pursuing endless growth is not sustainable.

> But when risk appetite is high and the company can afford to throw money at experiments, it makes sense to still pay for the rest of the 80%.

Does it, though? You're paying for full-time engineers to work on bets that might be completely misaligned even if they succeed, which also might be unlikely. And this makes up the majority of your company? It sounds ludicrous in any market.

How does a team of 30 not measure the usage of their system?

req/sec can be still high, it's just that the downstream service decides not to use the data

When the data your team serves is copied from service to service to provide high availability and no centralized failure point.

Can’t speak for Amazon, but when I was Google a bunch of high level leads were building a less generic / powerful / expressive Angular, implemented using Angular, so it’s an order of magnitude more sluggish. They were from eng prod (building tools for developers) but had no front end experience.

I’ve been around the block enough to see how management can tie BS projects back into the scope of some director level “vision”. When it’s a bull market, everyone in the org wins.

Yeah, that sounds like something that would happen at Google, and done by backend engineers

And then this one project adopts it for some random reason and you're left holding the bag

From what I've heard frontend engineers are considered less skilled than backend engineers at Google?

That's the general thinking across the industry not just at Google.

I don't think so?

From my unscientific observations from 3+ years ago: frontend has a lower bar of entry, and more competition by being the default software engineering discipline of choice for those transitioning from Design roles or coding boot camps.

It also empirically paid less, all things being equal.

Alexa has apparently torched 10 billion dollars (just in 2022!) with nothing to show for positive cash flow. It's a cool gadget but they bet the damn farm on it. Thousands of people working on it over the years, all to just lose billions of dollars.

I’d really like to understand what all those people are doing. Like I can understand hundreds, but thousands? Does this thing carry on normal conversations with this people at this point? Like does it do much more than act as an alarm clock or Amazon order taker?

Also $10 billion? Do they hemorrhage money on hardware?

Related experience: I spent two years on Google Assistant's media team. We were ~35 people. Projects included integration with YouTube (5 person team), the visual UI for touch screens in cars (8), performance improvements (3-4), making "play" and "pause" commands work through the lock screen (3 people), backend migration to a new struct for representing a user query built by the linguists (5 people)

Ah I’m starting to see how that can grow rapidly then if people are assigned dedicated projects instead of shifting around.

If people shift around, you still need people to support the previous work too, so you don't really gain anything. You still need 5 people to support and iterate features that YouTube effort etc.

> Ah I’m starting to see how that can grow rapidly then if people are assigned dedicated projects instead of shifting around.

You're thinking of Siri.

It sounds like that's the fat right there! I know it's an example, but 8 people for visual UI? That's an example of something that could be down by 1 person or 100 people, depending on your attitude of design by committee. Easy to imagine meetings about meetings studying 500 possible interaction schemes before honing in on a final 15 to run mock-ups to evaluate before deciding to etc etc.

Rube Goldberg vs just empowering someone to design something and move on.

So 1 person to design, prototype, and test the thing? The design has to go thru many iterations, feedback sessions, approvals by managers and their managers, approvals from the design system folks, approval from the engineers who have to implement the thing. Then the prototyping- should a designer also know Java and the Android SDK to build a prototype? And should they commit the code into production and set up an experiment for A/B testing? (Whole set of data collection and bias concerns and approvals needed here) Or if they build a prototype in, let's say, Figma, is that high fidelity enough? (More internal feedback and managers required) Then the testing: building a research plan that avoids bias, asks the right questions, defines how success is measured, yeah the feedback might be subjective but the process needs to be empirical. And then gathering all of this data and compiling a whole set of research findings and suggestions so the manager-of-manager whoever is in charge can feel comfortable with going ahead and actually building the thing.