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

the only reason us wrinkly brains can make so much money is because of the smooth brains, if the smooth brains succumb to the existential threat that AI poses who is going to spend the money that pays your obscene salary?

Looking down on normal people for being concerned about their future shows that you’re not just devoid of the most basic of human emotions, empathy, but you’re also far more smooth-brained than a sandwich artist. Even the lowly sandwich artist understands that they need their customers to have money.


I have extensive support experience, you're right, support is not easy, that said, there is a very easy win: hire enthusiastic support staff that are overqualified. Customers don't like slow support but slow support is forgivable if it is good support. Framework's support as described in the post is a disaster, the worst type of support. Bad support compounds: if your support agents are doing the bare minimum to move a ticket along, they're pissing off customers and missing insight that would be valuable to the teams responsible. A customer support team of 2 technical people will deliver more for the company than a customer support team of 10 untrained people.

The quick fix for any company in this situation is to require that technical people across the company take responsibility for customer support. Of course, they don't want to, hardware designers and software engineers can't imagine anything worse, but they've got to be exposed to how real customers are experiencing their work to have any hope of doing a good job. The exposure can be direct (in a crisis) or indirect (through technical support staff relaying).

If I were the Framework CEO, I'd be taking over customer support myself. A nightmare, but necessary to understand why their support is so bad.


I have been a chief developer of a large service (think millions of daily users) and I had a habit of opening the support inbox and replying to users. In a typical evening hour I would clear the inbox and successfully resolve a mountain of tickets. Not to mention the important insight I was collecting from real users' feedback and real bug reports.

And this is how it should be: an enthusiastic developer can churn support tickets in their free time much much faster than dedicated support staff does full time.


This is why everyone in our company does a bit of helpdesk work from time to time - seeing the issues at the coal face keeps things grounded.

You're doing the lord's work

> The quick fix for any company in this situation is to require that technical people across the company take responsibility for customer support.

I like your idea but I can also imagine that quite a few technical people will not have the necessary communication skills for being in customer support. They will need additional training if you want to unleash them onto your paying customers.


That’s ok, I’ve done a thing where they pick up a 1 or 2 hour shift and look through the tickets that have high message count so or have been open too long. They get paired up with a support tech.

‘Oh, yeah we have been seeing this bug, tell the customer to flash this version.’

That’s a known issue and the fix is going through Qa

Sounds like there was an issue with that part, rma.

The amount of tickets your pros can clear while skipping the script is insane.

Additionally, it can turn up or handle bugs that aren’t handled by the script.

We all read the diagnosis on this post and knew the part was bad about 3 minutes in and it’s not even our project.


> Bad support compounds

Louder for the people on the back

In the same way a bad medical diagnosis makes things harder later on, or bad legal advice, etc

But especially in the "silly" cases, this is extra disappointing. The case in TFA is not "you're holding it wrong" or "weird technical bug that happens once in a very while", it's a very clear issue of "should do A and it doesn't"


This ^

You're rich enough that you don't have to think about what you'll be working on a decade from now, or even a year from now. Anyone saying that building a software factory is the future of software engineering is looking days, perhaps weeks, into the future. Software factories are a meme that'll be dead in a month. Saying software engineers concerned about their career should learn to make software factories is hollow.

> You're rich enough that you don't have to think about what you'll be working on a decade from now, or even a year from now.

Perhaps. But I work right now, and I actually care about people other than myself. I would like to see the industry that made me successful keep making others successful.

We'll have to wait a month or two I suppose to find out if you're right, but I think all companies will be moving to the model of software factories. And It's not something I thought of recently -- I've been building a software factory with AI for a year now, and working with others who are building entire businesses on it.

And FWIW, pretty much every tech company is now moving towards the same model.


Maybe I am off but software factories isn't what we are doing per se. I am not sure how successful we are with AI (yet) but our token cost vs results is very exciting.

This is my first reading about your work but I like your attitude. Databricks is also a good service.

From what I am seeing, those who embrace AI are embracing the 'computer' from the 80's. The landscape and reality is still being defined.

To the AI haters and doomers I say buy the largest house with cash you can to enable your retirement. The stress that people experience is the thought of their crushing debt or obligations that can be met with a 90% reduction in spend. Of course, that requires a lifestyle change that most are unwilling to make.


Are they factories tailored to the specific product you are creating? I feel like I get good results with a harness, and lessons from factories can be incorporated into models and harnesses as long as the lessons are general.

> Engineers are now more productive than ever. Increasing productivity drives salaries up.

No it doesn't. Salaries have been decreasing in real terms over the last decade despite productivity improving. There was a period of time in which productivity increased alongside salaries but that does not prove causation.


> Salaries have been decreasing in real terms over the last decade despite productivity improving.

Do you have a source for that?


The parent poster’s argument is not entirely true, because wages did outpace inflation for awhile around the pandemic. But it’s truthy. Wages have been sliding lately, and that sliding is the longer term trend.

https://www.marketplace.org/story/2026/09/04/wage-growth-can...

Obviously, the above is talking about labor generally, not just SWEs. But speaking as a professional computer scientist at a place with decent salaries and excellent benefits, my raises over the last 5 years have barely outpaced inflation. It’s pretty annoying to realize that I probably would have earned a lot more money had I NOT gone back to school for a PhD. Not that I would make a different decision given the choice again, but money is much more abstract in your 20s than it is in your 40s.


and then next week you'll need to learn how to build a software factory factory.

you need a abstract software factory for that.

fwiw this is reporting fully diluted value which is different from what Nasdaq reports, hence the discrepancy that will be seen across different reporting. Nasdaq is outstanding shares (13,086,672,593) multiplied by the price which is currently $98 = $1,279,156,812,603 market cap at the time of writing.

For as long as software engineering has existed, there have been prolific developers and there have been slow developers and every developer in between. Success has never been a product of volume. Some guy producing 30 "complex open source libraries" by hand in 2020 or with a "software factory" in 2026 doesn't mean anything. @sindresorhus published >1k packages long before code could be generated. Software is a means to an end, generating 30 "complex open source libraries" for no end changes nothing. If anything, these software factories are showing just how worthless most code is. Certainly, a "software factory" can be an accelerant towards building software that achieves some end, but it cannot magic an end out of nothing, it cannot make software valuable or useful, purpose must come from people.

You’re describing search as if only the interface matters. As a searcher, I want there to be a steady source of good information today, tomorrow and a decade from now. A great interface that leads to less information being published is not a good thing. Short term, sure, but long term, no, the power of being able to search the web has always been the diverse variety of information. Instant answers are only going to get worse as the incentives to publish information dry up.

> Those consequences are the reason that most database systems are built for either transactions or analytics, and not both. [24]

> [...]

> [24] People have tried to build one system that does both in the past (HTAP, or hybrid transactional/analytical processing), but those attempts generally didn't work out. Some newer systems seem to be bridging the gap, however. CedarDB, the commercial version of Umbra (a research system from TU Munich), runs transactions and analytics on just one engine and one copy of the data. Databricks' Lakebase keeps Postgres (a row store) as the engine for transactional queries, but its storage layer rewrites the same data into columnar formats in the background, allowing an analytical engine to query it.

The footnote says "those attempts generally didn't work out" but then mentions Lakebase and CedarDB as newer efforts. Can anyone speak to whether or not CedarDB, Lakebase (and any others?) are achieving the fabled transactions + analytics in one database? Should we expect, at some point in the near future, for transactions + analytics to be standard in all databases, or is combining them a niche / esoteric thing with many tradeoffs?


There are always tradeoffs. Databases are fundamentally about managing optimizations, and the question is whether those tradeoffs make sense for your workload.

HTAP has historically been niche because the cost of analytics on fresh data wasn't worth the squeeze (sooner or later the MBAs step in). Separating the OLTP and OLAP layers gives you better workload isolation and the ability to optimize each independently.

Databricks Lakebase/LTAP uses Postgres for transactions and lakehouse engines for analytics, with a shared storage layer. They also use Spark to build Postgres pages in parallel and write them directly to storage, avoiding the single-writer bottleneck when loading large volumes of data.

SingleStore takes a different approach: a distributed database based on MySQL with an in memory rowstore for transactions tightly coupled with a merger to columnstore for analytics.

I don't think HTAP DBs will become the default for every use case. But I do think the space will grow as we move towards agents and automated decision making and optimizations.


Agree with everything here. Workload isolation and independent scaling in particular are the reasons I never see HTAP work out in practice, even beyond expected loss from not using the more specialized solution. A very low latency storage handoff between two engines always seems like 80% of value with 20% of operational headache.

> HTAP has historically been niche because the cost of analytics on fresh data wasn't worth the squeeze (sooner or later the MBAs step in). Separating the OLTP and OLAP layers gives you better workload isolation and the ability to optimize each independently.

That feels like something HTAP databases could offer as a choice!

- Things you can always do

- Things you can do when optimizing for fresh data

- Things you can do when optimizing for analytics on old data

- Things you can do, but with poor performance

This feels like one of those cases where people choose to make software simple instead of, and at the cost of (not) making operations simple.


You are correct! And sadly the vast majority of people don't want to make choices or think about that tradeoff.

If it can't outperform pure OLTP and it can't outperform pure OLAP then there is no pit of success for such an HTAP system.

The unfortunate part is that the vast majority of software out there in the world would be served perfectly with an HTAP solution.

The fortunate part is that for a lot of problems you can just run Postgres and then when you need analytics you can still use Postgres, until you hit scaling issues (if ever).

I've never seen a project choose to solve OLAP and OLTP upfront and succeed.


The workloads are different. An almost -in-sync analytics replica is a decent tradeoff (copy, lag etc).

I'm working on one such solution right now,


I think there's never going to be a point where you can have the kind of performance you can get by optimising a datastructure for a 'fixed' dataset. Nor can you get the maximal performance updating data if you need to maintain a useful datastructure.

A simple case is a sorted list, you can sort it after every update, you can ignore the ordering and sort it as needed, or you can use a datastructure that is typically O(log(N)) at both updates and retrieval. Better datastructurs help, but they cannot remove the tradeoff entirely (in fact O(log(N)) inserts are kind of theoretically minimal, though a trie can do it in O(1) for fixed length datatypes).


There's a general trick which combines the time complexities.

In the case of a sorted list, it's keeping a sorted list, along with a growing small unsorted list that you will query or sort when needed.

This works for kernel timer trees, database storage, and indexes.

Keep your "fixed" data up to some checkpoint in an efficient form, and your "new" data since the checkpoint in a faster-updating, slower-querying form. When doing queries, read both and merge, and maybe sort parts of the new data on demand. The new data will be slower to query, but it should also be smaller.

At carefully chosen times, transfer data from the new form to the "fixed" form, either with an incremental update if you can, or by rewriting the fixed form entirely. It doesn't have to stop at two layers, and shouldn't if the fixed form is large, or if you want unlimited scale.

Choosing the right times and how much to update (the whole key range, intervals, predicate subsets, ertc) is an interesting algorithm design problem. When you get it just right, you can get excellent performance in fast-queryable compression and continuous fast-updates at the same time. (I'm building a database engine that does this in my spare time.)



Should also point out that there's a deamortized version of B-S (can avoid e.g. worst-case cascading merge in an LSM-tree) but it's not very well-known (basically replaces Bentley-Saxe's binary representation with Myers' skew-binary).

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

https://jeffe.cs.illinois.edu/teaching/datastructures/2011/n...


It is certainly possible. I built a highly optimized KV store, originally designed to add tags (key-value pairs) for the object store I invented a few years ago. This tagging system turned out to resemble a columnar store database, so I started using it to create relational tables.

As predicted, it was very good at performing analytic operations; but it also was very fast at traditional OLTP operations too. I benchmarked a large number of queries against other databases and in the majority of cases, it was actually faster. In some cases it was much faster. It was able to do all this using just one copy of the data without needing to create and maintain separate indexes.

There are several short demo videos on my YouTube channel (listed in my profile).


MS SQL Server lets you choose the table storage as row or column as needed along with clustered indexes and other things, but these are explicit choices:

https://learn.microsoft.com/en-us/sql/relational-databases/i...


Lakebase is just cheating, they superglued a transactional database to an analytical one.

That said, it works wonderfully and I love it.


There's also TiDB which has decent adoption. It allows for columnar replicas of transactional data and exposes both over the same MySQL-compatible interface.

I'm not familiar with your product, just the blog post + cell architecture documentation. Question: the documentation states that basins are assigned to a cell on creation and cannot move. Why choose to have explicit DNS records instead of wildcards with a subdomain per location? e.g: {basin}.us-east-1.s2.dev where *.us-east-1.s2.dev points to the cell? My best guess is that you prefer to have explicit architecture in case you need to move to a new cell for some reason, but maybe there was a different reason? The stuff around hitting the database on every query that necessitates a firewall seems like a huge risk vs. a wildcard CNAME which adds no additional burden.

Also, unsolicited advice but having your website, integration endpoints, status page and email under the same domain is a bad idea, a single point of failure. Especially a domain that is due to expire in just a few months. You should have a separate domain for infrastructure and a separate status page domain, each renewed for 10 years in advance (the maximum). e.g: s2.dev for website, s2status.com for status page and s2stream(s).com for your Basin URLs.

Thank you for the advice, the status page move makes a lot of sense! I don't think we have the bandwidth to separate the basin subdomains just now, but very helpful in terms of SPoF.

edit: oh, and the 10 year renewal, that's the top of the list.


The main two reasons: we have multiple cells per location (we want to keep cells decent in size to reduce failures) and a move between cells is something we have on the roadmap. I tried to find a wildcard solution at the beginning of building this, I just couldn't find a good one.

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

Search: