HN Simulatornew | past | comments | lists | submit | fusslo's commentslogin

We've stopped reviewing code for code's sake. We review code to keep the codebase clean and consistent.

Before agentic development, the cost was large for changes at the review stage. Often the person writing the code didn't have the right context or experience to make the best choice. So reviewing would often blow up a pull request. Which sucked. Eventually every engineer would have AI generate a PR. People would review, blow it up, and the engineer would copy/paste the feedback into AI to generate another PR.

But now, we use reviews to enforce design decisions and get feedback from the other people who couldn't be involved in the initial design (for example if a new feature interacts with a module owned by another team).

The actual nuts-and-bolts of how the code is written is mostly unimportant. It was a large mentality shift for the company

We spend LOTS of time on the design, verification, testing. The verification and testing is built first. Agents take the input of the problem, design document, and current codebase to generate code. It iterates by running the verification tools. Then it runs the testing tools.

The better we make the tools for AI, the better the product code is. We've automated gathering power readings, timing reports, wifi/ble sniffing, smoke tests, feature tests, corner cases, robots for pressing buttons, static analysis tools, unit tests, customer styled system tests... tools to emulate cell phones, servers, etc...


> We spend LOTS of time on the design, verification, testing.

sure, this is short-term because you still have people around who wrote (most of) the code and understand it

once you no longer have a good understanding of the codebase, you also don't know what to test and validate

AI is great at writing tons of tests; whether they're the _right_ tests is another question


> The actual nuts-and-bolts of how the code is written is mostly unimportant

I am struggling to come to terms with this personally. I was someone who really tried hard to make sure the nuts and bolts was good quality. Kept on top of language features, made sure not to use deprecated APIs that could expose security problems, etc

It wasn't just "to make the code more maintainable" or anything, those choices mattered, or felt like they did

Now it feels like everyone is saying "those choices don't matter, and never did"

Have I just been wasting my time and focusing on the wrong stuff for my whole career?


No they absolute did matter. They just matter a lot less now.

I'm going to ignore the using an actually broken API, and instead think about using a sub-optimal one.

Before, you tried hard to get it right first time, because it was a pain to change if it wasn't.

Now, that change is much, much easier. A task like "rip out this deprecated package and replace with a new one" used to be a LOT of effort but, if we're honest, not an awful lot of skill. Now its quick and easy.

Making the right product choices, or architectural decisions remains just as important, because they're still an absolute nightmare to change later, but the nuts and bolts much less so.


right. Making the right choices is really hard

One thing I am still seeing is we STILL can't admit when we're wrong. We refuse to believe that we cannot think of everything before the code gets into production.

Really it's some core teams who do not have a 'sustaining' budget. They call themselves Services R&D.

For example they made a messaging stack with encryption, serialization, addressing, routing, etc. But it has a huge flaw in that there is a maximum packet size. The design is such that an app cannot implement chunking on top of the messaging. So there is no way to transmit data greater than the packet size or do any kind of data streaming. THere are 3 products now who are mandated to use this messaging stack but can't fit into the packets

It seems like a culture of 'deliver and move on' rather than 'deliver, support, improve'. Not sure how to fix this problem


> I am struggling to come to terms with this personally. I was someone who really tried hard to make sure the nuts and bolts was good quality.

Think of assembler vs compilers for higher abstraction level languages. When you write in C or Python you trust the compiler to do the right thing and don't routinely review generated assembly let alone machine code.

It's the same here. You write your program in natural language and have the compiler (ai) make it happen. This is what it will converge eventually to.


Huh, I remember discourse with the CEO (I think) of Monumental on twitter a while ago. There was a frenzy around anti-ai art at the time. If I remember right, he replied to many comments that very skilled human artisans still did the final touches on the sculptures that were cnc carved rough. It really seemed like he was taking that criticism seriously

But now, "... our system turns a 3D model and a block of marble into a finished sculpture, with no human in the loop."

I'm not judging one way or the other. I have 0 skin in this debate. I just find situations like that interesting; like trying to douse a publicity fire while secretly making wildfire


>There was a frenzy around anti-ai art at the time.

There still is. Most people are extremely hostile to AI art. (Albeit I see this case as more like mass-production than AI art per se).


It would be really nice to be able to afford a marble statue. But, really thinking about it I would more than happy with a cast statue, if the casting could have as much detail and crisp edges as marble.

I started looking into modern concretes (for about 5 minutes), and there seems to be a bunch of modern options. I don't know how they compare to marble. I wonder if there is any literature out there giving detail on their properties

But then again, there is an artist who had a bronze sculpture cast in china for crazy cheap. I can't rewatch it now, I think it was under $20k https://www.youtube.com/watch?v=7CtmiWlT6Z4


i think it lets you read the article without having to copy/paste the url in archive.ph

oh-right-I'm-not-supposed-to-be-allowed-to-read-WP

I honestly skip paywall articles or just click the archive/wallhop link so frequently that I may have forgotten one is expected to pay for WashingtonPost...


What is the incentive to produce good journalism if it doesn’t help you sell your news?

The actual problem seems to be a transaction costs problem.

Estimate how much you think it costs to write a single article and divide that by how much each user would have to pay if e.g. a million people read it. It's pennies each, sometimes fractions of a penny. But now how do you allow those people to each pay around one cent without requiring them to do far more than one cent worth of signing up for something, and without being coerced to tie their reading habits to their government ID through the banks?

So instead they try to charge for subscriptions, which then falls apart because you don't want to read a hundred articles from one publication, you want to read one each from a hundred publications, but you also don't want to pay for a hundred subscriptions.

If you actually want to solve it you need a digital payments system with low friction and low transaction costs that allows the buyer to remain anonymous. Which ads have been the stand-in for since the actual thing is heavily deterred if not outright banned by existing regulations.


People are generally intrinsically incentivized to do a good job a things they care about independent of penny pinching. So your comment would imply that journalists don't really care about journalism, which is a take I would probably agree with for most - but it's quite cynical, and probably not what you intended.

Now that we're well off in the weeds, this is also prevalent in the "LLMs ruined coding for me" discussions.

If you're in it purely for the joy of code, you can still code to your hearts' content. If there's a profit or career motive, then it's different.

If you're producing journalism for the pure love of communicating your findings you don't have a profit motive. If you do have a profit motive, well, you'll need to secure a profit in addition to or in place of your love of the game.


Journalists want to eat too.

I don't have an answer just anecdotally I know we'd use UPS (not even usps) to ship classified documents when I worked at the DOE

USPS accidentally delivered a bunch of classified military documents to my parents house when I was a kid. Some government official came out to personally take them back and thanked my parents profusely for returning them.

A classified F-35 component was shipped using UPS and it somehow ended up in China

https://www.cnbc.com/2026/09/24/chinaf-f35-components-hong-k...


In the early 00s I worked in print production and we printed a ton of material for the DoE. Their own employees would give us classified documents to print into binders for presentations/meetings all the time. We were basically a Kinkos...

UPS also somehow delivered stealth F35 Parts directly to China... no espionage required

turn it on/off really fast

The relay clicking is a feature so you can tell the brightness audibly :)


I recently had a similar issue. But instead I decided to go with industrial controls. I bought a Contactor with a NC and NO connections. I put a float switch in series with the activator coil circuit. So when the float triggers: NC->O and NO-C, which turns off the water and turns on the pump.

Contactors are built to run giant industrial motors for years, nevermind my little 1/4hp pump.

The way I built it lets my mother bypass it by unplugging the pump from the control box and plugging it right into the wall outlet.

So, hopefully it lasts years. If it doesnt, anyone can bypass it


Every sump pump I've seen has a built-in float switch that works.

I need an alarm because I've had a few pumps wear out, and also because I've had the power go out long enough to flood.

Your solution automates something that shouldn't have a computer involved in automating.


I'd love some classification for support tickets where i work

whenever i submit a ticket (need hardware, problem with software, access, etc etc) I'm supposed to know what business unit im submitting to. There are 63 last time I checked. Some go to global IT. some go to local IT. some go to to procurement, etc etc. There's no resources on how to select the right one

I just guess and there is a dedicated person on every team who reassigns tickets to the right business unit.

nightmare.

an llm that parses my request and auto-populates in the back end would be cracker jack


I've learned it's not that uncommon. Even letting friends track you. I remember a thread on X where people in NYC were posting their friend's real time location and saying things like 'I can always tell when she had a bad day' because her icon was at a particular juice bar. It does seem smart, especially for 20-somethings in major cities. But posting your friends location isn't terribly smart

It's a thing.

I, personally, don't think I've ever trusted anyone (parents included) enough to let them track me. I wouldn't feel safer. But maybe some people have good and healthy relationships with their parents (not that forcing your child to be tracked is healthy, but if your child chooses to have that safeguard...).


> I've learned it's not that uncommon. Even letting friends track you.

This is actually somewhat old news. Snapchat pioneered this with the Snap Map, where essentially the whole purpose was to broadcast where you are to your friends.

That was so long ago now that even Snapchat is long last its peak relevance.


There was also Google Latitude way before that. I remember how excited my friends were to watch other friends move about the city. It did come in handy one time when a friend left his phone behind at a corner store and didn't realize until later.

I don’t get how posting their location in this instance matters. Unless the friend has a stalker, is a famous celebrity, or is operating under some really unusual threat model, what risk is there in the whole world knowing some random 20-something is at a random juice bar in NYC? They aren’t being targeted, this isn’t a controversial location, etc.

If a friend is posting a person's location, they likely post other information about that person. The aggregate information could allow the assembly of a much more complete picture of a person than is possible by observing a person at a random juice bar.

For example: are they married? Do they live alone? Who are some of their friends? Where do they work? What do they do at work? Have they recently lost/found/experienced something interesting? These are the kinds of details that can enable social engineering in digital and physical spaces. It's likely easier to lure a person into a dangerous situation if you reference minute details from their personal life.

Sometimes the threat model is "has $10k in a bank account" or "is young & attractive" or "practices a religion I don't like" or "it's Tuesday."


It's not their data to share.

It's data entrusted to that specific person with the social contract to not share that data (assuming no permission for sharing was given). I would argue the same way if someone posted my full address online for example, you're breaking the social contract by sharing data I entrusted you with.

Perhaps they didn't tell their friend about that one abusive ex or an affair their having. Additionally in this case it is telling people that when this person is in a certain location they are having a bad day, possibly opening them up to harassment a scam or manipulation.

There are a lot of people that will use that information to their advantage, even if it is just for trolling.

For me, if I found out about a post like this I would stop any contact with the person effective immediately after informing them of not appreciating the post.


well in my particular example it's a juicebar she frequents when she's had a bad day. So, she probably lives or works in the neighborhood. And a detail I didn't share was that she was attractive (her icon on the map was her headshot). So someone could get it in their head to approach the person and know what she looks like, where she might be, and believe she needs cheering up

And it's NYC which has all sorts of people, many of which you may not want to be approached by.

If the juicebar gets tagged, then the people working there may start interacting with the friend in a way she may not appreciate.

Sending out life specifics, even if your not famous or have a stalker can be detrimental. And, it wasn't the friend's choice either. So, the original poster may loose a friend in exchange for likes on X.


The thing that never makes sense about this threat model is she lives or works in this neighborhood, which in NYC is many thousands of people. They see her, they can approach her, they can follow her back to her apartment if they chose. To get to said juice bar, she’s likely walked past a dozen mentally ill homeless people and maybe ridden the subway, where you’re confined to a metal tube underground with NYC’s most interesting for minutes at a time.

But none of that registers as a risk: instead it’s a hypothetical X user who is so struck by a headshot on a map that they travel across the city on the off chance she’s still at the juice bar? That’s like worrying about being struck by a meteorite, but paying no attention to the risk of getting hit by a car.


why do you say that?

I didn't see it in the article nor the 2025 study https://www.nature.com/articles/s41893-025-01578-w

Additionally, the article points to Spain and Lake Constance in the Alps exhibiting similar effects.


Wow the Spain is even more strange. It was over farmed and plowed so much that soil is degraded to become like concrete and barren because the water doesn't soak in.

https://www.youtube.com/watch?v=ZUzZk_uqSWc&t=107s

For that type of devastation, I recall reading years ago that growing rye and then rolling it instead of plowing it and replanting rye again for multiple seasons can create moisture retaining top soil. They would also need to add ditches to have pools of water for wild life to make a come back.

It is a completely different phenomenon. In 500 years all these lands are going to have an ecosystem again even if they don't do anything but they have to live with it for 100s of years if they don't do anything.


I looked at few documents about japan and I believe it was conclusive.

https://www.nippon.com/en/articles/-/14855

I will look up spain when I get time and get back to you. I have to be somewhere now.


> We continue to reach out directly to those sites that we identify as particularly high impact and try to notify them, though this is happening on an admittedly small scale. The process of responsible disclosure consumes a lot of resources, use of security.txt is low, and the vast majority of our emails and messages go unanswered. It's simply not possible for us to scale this up.

...why?

Some of the sites collect credit card information. Wouldn't they have a 'contact us' or some other legal identity information required for collecting and charging cc's?

Other than looking at security.txt what else have they tried? did they try programmatically checking for a 'contact us'? or the old trick of emailing 'webmaster@domain.com' or 'admin@domain.com'? did I miss those details in the article?

Maybe things have changed in the past 10 years, but it was a MASSIVE deal if a website ended up on some site claiming it has malware. Like, contract-ending, people-getting-fired, 10s of thousands of dollars in lawyer fees kinda deal.

I don't think 'it's hard' is a good excuse to ignore responsible disclosure


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: