Is Meta destroying its engineering organization?

There are a lot of terrible practices out there in the world that you should stay wary of. Too many false positive alerts, flakey tests, not enough tests, not listening to users, taking a solution because it's easy and popular but not necessarily a good fit for your specific requirements, etc. Many of these practices are popular unfortunately. That's not to say others don't have great ideas, just don't copy them blindly.

None of those were acquisitions

Also very strong teams working on db/data, source control, performance, reliability, pl/compilers, etc.

As a user, Instagram is about the only part of the Meta empire that doesn't suck

There were other reasonably nicely engineered things there, like some of the code-as-configuration things in PHP like the ORM and the structured logging. Where you just write and commit code and the databases and data warehouses just get set up automatically.

The choice to have a monolithic web stack at Facebook had some benefits. The latency from landing a diff to being in prod was a couple of hours, not matter what part of the site you touched. And given an arbitrary URL, there was a tool that used static analysis to tell you which file handled that request. Compared to Google where every damn thing is a micro-service with its own variable release schedule and different ways of doing request routing. Trying to figure out where a request goes is combination of dynamic tracing of the RPC trace behind it plus a bunch of grepping through the related code bases.

All that said, I was far away from doing real engineering at Facebook. I accidentally joined a growth team. Which meant our teams remit was churning out all kinds of experiments to see what we could to do to make our team's number go up. I did not find this type of work enjoyable and quit. So I have a narrow perspective.

Also to give these multinational corporations some benefit of the doubt, some of the way they operate is path-dependent. They have hundreds of millions of lines code. So they can get to a point where it makes sense to hire a bunch of compiler engineers to fork PHP and make the Hack language. Whereas a startup would almost certainly never create a new programming language to write their product. To take solutions of this context and apply them to the outside world might not make sense.

[1]: https://engineering.fb.com/2020/03/02/data-infrastructure/me...

[2]: https://cgsql.dev/

That makes sense from a consumer point of view. I don't use FB much. In fact, I recently logged on a few times after a few years of a break. It's terrible. You can't even get conversation history anymore. You can talk to someone via the app on your phone, open Facebook in the browser on your computer and not see the messages there. Makes me wonder why is Meta still around?

I might not know him, but I don't think he changed. I think the circumstances changed and they happened to lead to another expression of who he is.

IMO the "AI psicosis" is just the latest expression of the hyper-capitalist ideology that has taken over Silicon Valley (and the USA).

Engineering at these companies was special, because there was good money to be made from software at practically nil materials and distribution costs. It was all labour; very specific labour. Instead of a million-dollar idea taking months and lots of $$$ to prototype, mass-produce, and distribute, the right people could type for a week, press a button, and reach millions of people pretty much instantly.

Of course capitalism is going to be kind on this type of labour. But as it becomes clear this labour can be replaced with cheaper automation, we just switch to see the other side of the coin; one employees in other industries have seen for a long time. As an employee, you're a human resource, a cost-center, and without proper labour protections, you get treated as such.

Mark Zuckerberg has also always been like this. Not with the engineers back when they were more special, but with the content moderators for example, who have long been considered "replaceable". plenty has been written about how terrible it is to work in content moderation at Meta, the quotas, the little thought given to doing a good job, to mental health, etc. As engineers keep getting less special, they'll keep seeing the face of Zuckerberg that exploits resources, decreases cost, and demands higher margins.

> I am flabbergasted that I am a lowly peon in the machine, and yet unless I'm missing something, I seem to be a lot smarter than most of the people I see running these companies.

This has also long been the case, especially in other industries. I can assure you an average barista at Starbucks knows more about making good coffee and keeping customers happy than the CEO.

Credit where credit is due, the CEO probably does know more about playing the politics, pleasing investors and how to ruthlessly optimize organizations.

I know that lines of codes and PRs go up, but how much of that actually moves the needle forward?

We already know what the algorithm does to normal people - it should be treated like a radioactive object by anyone in charge of anything. Very powerful and strictly used judiciously and in small doses. Instead we've got some of the most powerful people on earth just cooking their brains on this shit just like anyone else watching reels on the bus.

This stuff is putting an expiration date on your domain experience, and leadership is salivating at the chance to cut OpEx.

At bare minimum, every PM is soon going to be able to do your work. That's a huge talent pool to draw from. This also opens the door to easy offshoring.

They'll use layoffs to get rid of the existing high salary earners and backfill with new hires earning 70%, then 50%, then...

I can't believe how many of you are absolutely certain you're going to be working in software in ten years. Let alone five.

Sometimes they do know the work being done and how it makes them money. But they want MOORE and having engineers putting dampers on their latest wishes (this will break prod, what are the requirements,...) is something they truly dislike.

It was the engineers at the ground floor who I watched become religious about it. They have been the ones pushing for deeper and deeper tooling. They have been the ones convincing leadership that this is the future and so now, well, it's leadership who is saying, we want more adoption.

This psychosis is happening at every level in our industry, and this isn't a big tech, or leadership problem. It's all our problems.

Something that can be useful but being over-reliant on especially at a management level, could be disastrous.

For a coding agent, for example, there is *very detailed* analysis of the turns and ranking of different portions of the conversation.

Adherence or deviation from specific rules matters. Writing quality matters. Expertise in the topic under discussion matters. Having intuition for the tone and beat of a good conversation matters.

Scoring a 15-20 turn conversation can easily take two and a half hours.

Clicking submit does not mean the author is done. Many annotations will be turned back to them by a reviewer to touch up in some way.

This work can be far more mentally taxing than programming, is measured much more by completions more of a timed exercise than SWE.

FWIW, Meta employees would probably make great coding agent conversation annotators. But it is absolutely not SWE and they won't enjoy it (for long).

I doubt it'll pay off. Let's face it, the average FB engineer is no better than the average Polish engineer, and the Polish guy is 20x cheaper. If you wanted cost-effective labeling you'd send it to Poland. Or better, Chile.

Gavin from Silicon Valley did it first

I haven't interviewed with them in almost 10 years. But aren't they doing the same interview everyone else does?

> Forced data labeling with 4,500+ engineers is to generate high-quality RLHF

I doubt that you get high quality from forced reassignments where the now-data labelers don’t actually want to do that kind of work.

It’s crazy to think that Meta leadership believed that it makes sense.

It’s a matter of time before the regulatory hammer falls on them and they’re hemmed in via civil courts due to damages. Many of the civil suits they’re losing are petty weak cases showing juries and judges literally hate meta. These are signs that their revenues are short term money grabs due to blindly chasing iRev at the cost of literally everything else.

Even hacking they only really follow through on if their large business credit extensions are at risk or chargebacks. Debit and direct funding isn’t pursued because they’re not loss liable - if it’s literal theft via debit cards, it’s not a priority. If it’s their own money via credit that’s refundable, they’re all hands on deck.

This will all blow back on them at some point and that point isn’t far aware. Courts, governments, banks, they’re all starting to notice the lawlessness and pure avarice. The consequences will severely impair their ads business, require them to undergo worse consent, oversight, and audit that waving an AI hand at won’t be sufficient, especially when the courts are concerned as judges are especially skeptical of AI solutions to court orders.

Their only out is to find another business model, which they’ve been trying to do without success since Facebook was first launched.

That's because your engineering career coincides with a race to the bottom, where advertising-saturated, FOMO-afflicted consumers demonstrated a preference for accumulating as many cheap/free/subsidized things that they could over a few durable, valuable things that genuinely benefit them.

It wasn't always that way, and if the economy does encounter a strong correction, it could very well change again.

There's always the option of getting rid of all the engineers working on new stuff, and having a small support staff. Often times, customers would even prefer that.

So, let's see, the top tech companies in revenue are Google, Apple, Meta, Amazon, Microsoft, and some of the fastest growing ones are OpenAI and Anthropic.

Do you know what all these have in common? They give extremely high compensation even compared to other large companies (microsoft is a bit of an exception here).

So you think they just do this out of the goodness of their hearts, that these kind CEOs who would lay off tons of people on a whim, don't think engineering matters, but are paying 300k, 400k, 600k, 800k to software engineers?

With just a little bit of hyperbole:

The culture of Google is that you hire "the best" developers (say top 0.1%) and hamstring them with process and cumbersome tools so that you need 10x as many of them as another company would need and pay them 3x market rates, but it is OK because (1) at the scale they work at they can amortize the cost over a large user base and (2) they make monopoly products. Google's systems are highly scalable, I grant that, but they have the first mover disadvantage that their foundations are first-generation and not based on experience and still slowing them down... but the market can't discipline them.

Facebook on the other hand, cares about internal DX, sees it a problem when developers are stuck with cumbersome processes, and greases the skids. They benefit from huge scale and monopoly profits but Zuck is keeping more in his pocket than he would be if he did things like Google.

The problem with React is that mediocre developers can use it to build big things that are too big for them to handle. The problem with Kubernetes is that above-average developers can use it to build small things they can't handle. And the scalability of Kube is more than almost all of the industry needs. That is, a system like what Hazelcast was before it became an analytics play could support clusters of 30 or so big nodes (two racks) and there are probably just a few 100 systems in the worlds that really need to get bigger than that.

Rather than two competitive products I think React and Kube are both market dominant products that reveal the engineering culture, values and impact of the two organizations.

I can say that when I first started with React I thought it sucked and looked at Vue and Svelte and similar things. My take on all of them was that they made the internal frameworks I was using to build very complex RIAs in the 2005-2010 time frame look like something that fell off a UFO.

I first thought Vue reflected the way I thought about conventional "webby" applications particularly in terms of treating lists as a first-class object. I eventually learned with React how to draw absolutely anything, even whole 3-d worlds!

https://aframe.io/

I guess what I'll say about Svelte is "if you make such a radical change in your framework did you really believe in it?" Early on I stuck to controlled forms in React because uncontrolled forms seemed structurally unstable (add one too many features and it all breaks) and then I discovered

https://react-hook-form.com/

which is easy and very high performing. What I really like about React is not so much the DX or UX but rather the way it uses functions to seemingly transform a language into something else in a conceptually simple way... And how it solves the problem of composing components written by different people and organizations to the extent that we've now got the terrible problem of managing an app that has 50 third party components and 5 CSS management frameworks.

Specifically I wrote a bunch of React components for making little biosignals applications that can (say) show two people's heartrate from bluetooth LE and show my breathing based on a strap i am wearing and another person's breathing based on a $20 radar from China.

I can pretty easily snap together the components and the system that feeds the state to the components by writing code. It works great, it's not that hard to do, it looks great.

But: I really wish I could make something where I could drag and drop display and data acquisition and processing components like LabView. Actually I know a lot about how to do the dynamic processing (Hint: read the Dragon book, not On Lisp) but React doesn't support dynamically assembled components... But I know Javascript systems can because I was writing them 2005-2010 back when browsers didn't have async and all the great affordances they offer now.

I built my own action framework that gives me the ability to use Jotai getters to read atom data, launch asynchronous javascript, and then write to atom data via Jotai setters without ever having to fuss with useEffect myself. Jotai just handles the messy state transition work. My components used to be a jumble of DOM event handler, business logic, and markup, and now the business logic is all extracted to the separate action components.

React makes it hard to test business logic in isolation, and I am hoping my action framework could do a better job of that.

I think you might have Reid Hoffman (founder of LinkedIn) mixed up with Reed Hastings (founder of Netflix). :)

it's less than two years, probably well below average. maybe it was more previously

That's the thing right

So I was research so both metaverse and AI adjacent. FAIR was industry leading, just not in the sexy field of LLM. FAIR was passed around like a joint at a student house, finally landing under Cox for "product". But FAIR is a research org, so it was a bad fit, run by an even worse leader. (why can't you deliver a new industry leading LLM in 5 weeks? or some other stupid shit)

The metaverse flopped because there was no clear leadership over features/user experience/hardware. Hardware has a 2-4 year lead time. This means that you need to plan your features 5 years in advance. The average horizon for any software feature in oculus was like 4 months.

Because of the huge influx of non game/graphics/hardware engineers the same mistake about "oh lets build a x but for oculus" happened every year. When I left they had contracted a company to re-make unity but for horizon. At the same time they were also making a blender clone, but in react or some stupid shitty idea.

At no point was there a comprehensive plan for what the UX should be like. there were lots of plans that people made, posted about, got many likes. Lots of redesigns of the button, new social features, avatars etc.

Carmak kept on banging on about time to fun, but he never managed to actually make that work. So a social company with a massive social graph, has a product where you can't easily join your friends in a game. (that might of changed, they revoked all my games when I left so I haven't logged in.)

But your point right, in all the years threads is the only new product they have launched, and that only happend by accident.

Meta's SOP is basically have an idea that gets zuck hard, do a small PoC, it shows promise, scale the team from 10 to 3000, and don't deliver anything.

The labelling I don't think is an attrition thing, I think its doing standard facebook shit, throw people at the problem, without thinking about how that would work

There are other reasons one might reasonably object to keystroke tracking, of course.

I don't even like doing stuff like this on my phone.

So, yeah, people do absolutely braindead shit with their company-furnished equipment. It's fucking mind-boggling.

I work to put food in the table.

Turned out he was schtupping the vp of the design company (his wife told the admin assistant during the divorce)

Also the genocides.

Can you expand on this? Aren't there plenty of "not-amazing-but-definitely-not-evil" organizations out there which need talented engineers?

People created PyTorch and React, they happen to be working at Meta at the time.

Maybe it's unlikely they'd create those working elsewhere, but I think it's much more unlikely that someone else at Meta would have created the same thing without those people there.

Oh yes, I would color Meta.

https://en.wikipedia.org/wiki/Facebook_emotional_manipulatio...

"Company over country!" -- Mark Zuckerberg https://www.yahoo.com/news/book-zuckerberg-called-company-ov...

The advertisements within Instagram for Threads almost always seem to be fairly thinly disguised engagement & rage bait. Every time it gets me, I feel an increasing desire to move away from the whole ecosystem.

I don't understand this either, there are so many clearly advantageous ideas and experiments to be be carried out, that can make discussions better, thinking clearer and help people actually connect. But instead they're only thinking about how to optimize the ad-machine in the end, so depressing to see.

Facebook seems to be the most misinformed audience - an LLM fact checker would be a great addition.

  The chairman of the U.N. Independent International Fact-Finding Mission on Myanmar stated that Facebook played a "determining role" in the Rohingya genocide.[98] Facebook has been accused of enabling the spread of Islamophobic content which targets the Rohingya people.[99] The United Nations Human Rights Council has called the platform "a useful instrument for those seeking to spread hate".[100]

True, none of us are innocent.

> When you step back and try to explain exactly what the company has done that's so bad, there's nothing there.

WTF?

The reality is that more engineers need to be able to grow a spine, have longer term thinking and actually stand their ground when it comes to these companies. You could not pay me enough money to work at Meta or Palantir and while it's true there are very few ethical or noble companies, working for Meta is more akin to completely throwing out your ethical compass.

And BMW meetups are not good ways to meet high-salary people. Likely the exact opposite.

Not OP but I can say with 100% truth and certainty that it wouldn’t matter how much money they offered - I would not work for meta. Some things matter more than money.

I feel gross about the place I’m at and I don’t want to lose that feeling.

FYI there's nothing that said the depression Facebook intentionally causes in teenagers is limited to just girls.

You really think if they didn't work there, someone else wouldn't?

You really think them and only them are the people capable of doing whatever technical things are causing the problems you perceive?

And while it's true that many organizations carry dirty laundry, especially as they scale into larger organizations with fingers in more pies, Meta's business model is specifically to maximize engagement by any means available so that it can sell their eyeballs to the most lucrative advertising opportunities.

In Facebook's early days and as Whatsapp continues to do, their products may incidentally provided a useful societal function for earnestly connecting people in the way that those people wanted to be connected. But there's no way to look at it through the lens of a socially responsible business -- a qualitative difference from an organization simply not having "clean hands".

people say this as if having principles will leave you jobless inevitably, but there is so much evidence to the contrary that this rings more hollow the older I get.

I agree with your overall concept of empathy towards others, I disagree with the premise that all organizations are unethical and that there isn't room for the morally principal'd and employed in this world.

contrarily I believe that a morally principal'd and skilled engineer is so rare in this world that there are a few organizations that would snap up every single one they could find if the network was there to find them.

There are plenty of organizations that don't enable genocide.

Either they are actively a narcissist, actively immoral, not intelligent enough to understand the vibe there, or they're actually unhappy and not being honest enough to tell you they hate it there. I have a hard time envisioning any other possibility. The place actively filters out morally coherent and intelligent people.

Yes I was there.

This seems like a fancily dressed up "fuck you, got mine"