Hacker News Reader

Comment by eru | Hacker News Reader

On the scale of a company, augmenting is replacing. If a worker plus AI can do the work of two workers without AI (but cheaper), you go for that; and it doesn't matter how good or bad AI is without the human.

View on HN

The point is if a worker plus AI can do the work of two workers without AI, then why not keep both workers and have them both use AI to have the equivalent of four non-assisted workers?

Because you don’t have enough work that really needs doing, at least in that particular area. You cut engineers because the bottleneck to increased revenue isn’t software features or bugs, it’s marketing/sales; human beings’ limited attention for which there is now more competition than ever; and customers’ available funds.

ETA: this is sometimes (though not always) very different for a mature company than an early stage startup.

That is a very convenient message for marketing and sales people. The fact that their whole job is crafting messages shouldn’t raise any eyebrows.

Ha! I’d never thought about it like that but…yeah.

I suspect another big part of it is that marketing and sales are relatively easy to measure and to scale.

You can hire one, two, or three new salespeople and expect that revenue will change more or less proportionately. Fixing (or ignoring) a handful issues doesn’t scale so smoothly—-there are jumps where the product suddenly seems much better/worse.

I took 'engineer' vs 'marketing' and 'sales' as just examples in the grandfather message. The essence was about bottlenecks to revenue.

Because the entire structure of the business is designed for approximately the amount of work it currently does and likely has no particular immediate use for twice as much work in most departments.

In 20 years I have never been on a team that didn't have twice as much work as we had people to do it.

Businesses are not magically efficient

Your experience of how the world works is usually because of what work you have done. You can't grow twice as many crops, sell twice as many groceries, drive twice as many busses, because of AI - fundamentally there's a consumption problem as well.

Many businesses are not bottlenecked by processes that are computer based.

But the firms in the headlines doing layoffs after layoffs aren't growing crops, selling groceries, or driving busses... They're knowledge work roles in companies selling intangible products and services. It's large corporations doing this much more than SMBs.

And I’m sure those companies also have “backlogs" due to limited labor/labor costs. There are always shelves to face, vehicles with deferred maintenance, and so on.

Obviously, there are limits: I’m not sure what my local grocery store or bus line would do with 100 new workers, but I have no doubt they could put a few people to work right away.

Well, but they aren't doing so right now, because labour has costs.

If your existing labour suddenly gets more productive, there's still a trade-off between cost and benefit to be made. And it would be a great coincidence if the optimal trade-off were exactly at the same headcount as the old trade-off under the old productivity figures.

It's more likely that the optimal trade-off for the business under the new conditions is at some other headcount.

They’re also the ones constantly hiring and recruiting because internally nearly everyone benefits to having more people “under them”, and there’s a massive HR/Talent team that doesn’t go into hibernation after a 20% workforce reduction. Organizations want to grow, not because they need to but because it’s in the best interest of nearly all individuals still on the inside.

That's why it's so important that people get some stock compensation, so that when the stock goes up when they finally fire people the incentives are aligned.

Diminishing returns on additional labor.

There's only so much to do and coordination costs (already burdensome) become overwhelming.

I think you're viewing this from the perspective of someone who has a functioning brain and plentiful concepts and ideas that aren't being built because you're labour-constrained. Companies like Meta simply don't have productive uses for all of that human + AI labour. Meta spends tens of billions a year paying people to throw shit at the wall and see what sticks. If the idea well you're going to is running dry, AI with a smaller number of humans can slop out the stuff you do want to build more efficiently is their implicit argument, especially when you don't care about quality (as is the case with Meta). Layoffs are also being used to tell a story around efficiency to investors while companies wait for the billions they're plowing in AI actually show profit.

The demand doesn't necessarily double.

Except that comforting C-suite narrative does not reflect reality. 2026 agents both increase productivity by knocking clearly specified but error-prone and tedious tasks out of the park whilst simultaneously vexing and annoying their users with hallucinations and downright lies on tasks with intrinsic ambiguity. This is made worse by the token providers with their constant tweaks to their deployments to cut costs w/o losing accuracy which flat doesn't work out well for the end user.

If one woman can produce a baby on 9 months, why can't you get 9 women pregnant and produce a baby in one month?

That works for throughput just fine. And if you have a bit of an inventory, you can hide the latency.

That's also how CPUs are made and sold: it takes a few months to make a physical CPU, but you can just go into a shop and buy one whenever you feel like it.

I think there are risks:

- AI pricing is variable, probably the cheapest it will ever be right now

- AI produces a lot more shit for humans to review, and you will always need humans. If you don’t focus on keeping things simple you will probably play yourself unless you’re good at separating out blast radiuses.

- I see a lot of super low quality work that doesn’t solve the problem but it’s like look that guy solved the problem in one day! Promote him! Everyone is happy except for the end users who for whatever reason are being totally ignored (whose problem it fails to appropriately solve) and I saw this in accounting software so…hello eventual lawsuits?

Why wouldn’t inference just keep getting better and cheaper as hardware and algorithms improve?

The typical playbook for a VC funded startup is to race to a monopoly where the company can have higher margins. Prices continuing to go down for the consumer over time would require competition to stay high in the long term, and even then it’s not clear if even current prices are profitable.

The current level of AI has plenty of inherent competition from local models. In the long term, most of the profit will probably be from very smart models that run at something closer to datacenter scale over long inference loops - where local inference can't do much and even third-party inference/small neoclouds will be severely challenged. That is a very natural "moat" and has natural cross-efficiencies with AI model training, which requires a similar scale.

AI is definitely not the cheapest it will ever be right now. The frontier is getting more expensive, but the same capability will get cheaper over time.

That's not always the case. Augmenting certainly can mean that, but it can also mean doing something that people couldn't do before.

For example, looking through meta data in a SQL environment that you didn't know existed to troubleshoot an issue. And a million other things. The odds of any employee not knowing everything are very good, even when humanity as a whole had already discovered that thing.

When the AI models hallucinate up a catastrophe, managers will reevaluate that calculus.

Humans are accountable and act accordingly, models are not.

Yeah, it's probably be something like:

2025: agents are the future!

2026: we don't need any new employees

_stuff gets real_

2027: wow, employees are actually pretty good value.*

Not all humans act like they are accountable.

Reply on news.ycombinator.com