Anyway, I dispute that there's "zero switching" costs to go from proprietary keyboard shortcuts to english sentences as the interface.
In the bloomberg example, the shortcuts are precise. LLM's responses are not always what you want.
Imagine being a vim or emacs user and have those replaced by something you have to type entire sentences for functionality.
"Public Data Access → Commoditized"
Also, no. Today I tried to have gemini pro give me some data from wikipedia for a list of countries that I supplied. It gave me data and source links, but the numbers were all wrong! I would have zero confidence in this for anything serious.
The bottom of the market fell out because of political turmoil. The U.S just revised job numbers down by a million for the first time in decades. The author’s thesis about the future of software businesses could be right or wrong but it is not the cause of the recent market collapse.
> Now consider what the same analyst does with an LLM agent:
"Show me all software companies with over $1B market cap, P/E under 30, and revenue growing over 20% year over year. Build a DCF model for the top 5. Run sensitivity analysis on discount rate and terminal growth."
While I think LLMs can improve the interface and help users learn/generate domain specific languages, I don’t see how a professional can trust an llm to get a technical request like this correct without verification. Wouldn’t a financial professional trust the Bloomberg llm agent that translates their request into a set of Bloomberg commands more?
I find this line of thinking so hard to follow. I happen to have worked at FactSet and when I was there we were stuck supporting chromium 18 because it was embedded in the version of the workstation that a partner had signed a contract which prevented them from needing to upgrade the base install. SASS is of course to some extent about functionality. Maybe it will make sense for larger customers to duplicate functionality themselves using AI. I very much doubt we’re anywhere near being able to prompt Claude and get FactSet out but fine we can grant that. The reason SASS is so valuable though is because it allows firms to pass off the costs and risk of maintenance behind SLAs. Why would you want to bring that risk in-house when you can essentially buy insurance against software errors by paying a SASS company?
I can follow the arguments, and I find many of them plausible. But LLMs are still unreliable and require attention and verification. Ultimately, it's an economic question: the cost of training the model and the computing power required to produce accurate results.
The strongest argument is the one about the interface. LLMs will definitely have a large impact. But under the hood, I still expect to see a lot of formally verified code, written by engineers with domain knowledge, with support by AI.
It’s worth asking: what do Wall Street traders know about building software companies? Almost nothing. Anyone who has attempted to start a startup knows that the software is always the easy part. Building the business is hard. The notion that we’re going to undo 100K+ years of specialization just so that companies can run mediocre, buggy versions of SaaS tools just to save a few bucks is crazy to me.
SaaS stocks are currently the buying opportunity of a lifetime.
My experience is that in most of small companies, the revenue extraction cycle from big SaaS suites has gone far enough that people are moving to self-hosted (or rather colocated with a local-ish provider that installs and provides unofficial support) software, like they used to do before the great plague.
Considering how Zipf's law works, there might be a huge discrepancy coming on as we see the products deteriorating from all the AI and H1B spaghetti code to the point where LibreOffice appears quite competent by comparison. Most of the people I worked with just want to sell lamps or furniture or trumpets or whatever they do, and the inventions of modern SaaS make this a lot harder to do. Once enough small businesses stop paying their 5-user subscriptions, I think this whole thing will pivot heavily into the favor of those that just maintained their product well and didn't ensloppify it in the mantime.
I think this is more interesting as a rubric than as a prediction. I agree with some of it and not with others; I don't know if we're "cooked" or not; I do like how they've broken vertical software's moats down though.
1. I don't buy that chat interfaces will replacing existing user interfaces. I'm in particular a little bit familiar with Bloomberg's user culture. I don't know that I buy that it's going to be replaced with LLM chat prompts. But software agents are going to make faithfully reproducing those existing user interfaces much easier, so: half credit?
2. Half credit again on LLMs vaporizing the "business logic" moat, because the vertical-specific rules that justified the original software market are I think a lot harder to encode in Markdown than the 1 week they gave it, and also verification becomes a bear as more ground-truth business logic is replaced with nondeterministic AI output. There's a thing happening here for sure, I just don't buy it's as decisive as they say.
3. Public data access: I 100% buy this. If this was a real moat, it's dead.
4. Talent scarcity: same deal. Remember, we're talking about vertical software, where the underlying technical work is fairly repetitive and best-practices driven; it's the exact slice of software development work LLMs excel at.
5. Bundling (you get IB messaging along with your charting and your news service); maybe. This point feels tautological. Work out what LLMs do to each of the bundled experiences and there's your answer for how resilient that moat is.
6. Proprietary data: I think they're just dead on right here, and it does indeed seem to be a good time to be a company like Bloomberg?
7. Regs lock-in: half credit, because AI does make regs compliance a lot easier, and I think we're at the very early stages of seeing how.
8. Network effects seems like a repetition of "bundling" and if I have a qualm about this rubric it's that they made it look like an even 10, so they could have clean wins and losses.
9. Transaction embedding (ie, being a payment processor or a loan originator) also seems tautological; it's a moat, sure, but they're begging the question of whether AI enables people to stand up viable competitors.
10. I think "system of record" and "transaction embedding" are kind of the same moat.
I wish people would not blog on X (I will call it X when it's used as a crappy blog platform); these ASCII charts are awful. But that's neither here nor &c.
> Please don't use HN primarily for promotion. It's ok to post your own stuff part of the time, but the primary use of the site should be for curiosity.
18 comments
[ 2.6 ms ] story [ 37.3 ms ] threadAnyway, I dispute that there's "zero switching" costs to go from proprietary keyboard shortcuts to english sentences as the interface.
In the bloomberg example, the shortcuts are precise. LLM's responses are not always what you want.
Imagine being a vim or emacs user and have those replaced by something you have to type entire sentences for functionality.
"Public Data Access → Commoditized"
Also, no. Today I tried to have gemini pro give me some data from wikipedia for a list of countries that I supplied. It gave me data and source links, but the numbers were all wrong! I would have zero confidence in this for anything serious.
The overall index has been pretty well flat. What sectors gained?
And surely there aren't 140 "software and services" companies in the top 500 by market cap?
While I think LLMs can improve the interface and help users learn/generate domain specific languages, I don’t see how a professional can trust an llm to get a technical request like this correct without verification. Wouldn’t a financial professional trust the Bloomberg llm agent that translates their request into a set of Bloomberg commands more?
The strongest argument is the one about the interface. LLMs will definitely have a large impact. But under the hood, I still expect to see a lot of formally verified code, written by engineers with domain knowledge, with support by AI.
Its another ad, no?
SaaS stocks are currently the buying opportunity of a lifetime.
Is anything down far enough to look like the opportunity of a lifetime right now?
There were some deals out there over the last few years. I still kick myself on a daily basis for not taking Carvana at $5 in 2023.
Considering how Zipf's law works, there might be a huge discrepancy coming on as we see the products deteriorating from all the AI and H1B spaghetti code to the point where LibreOffice appears quite competent by comparison. Most of the people I worked with just want to sell lamps or furniture or trumpets or whatever they do, and the inventions of modern SaaS make this a lot harder to do. Once enough small businesses stop paying their 5-user subscriptions, I think this whole thing will pivot heavily into the favor of those that just maintained their product well and didn't ensloppify it in the mantime.
1. I don't buy that chat interfaces will replacing existing user interfaces. I'm in particular a little bit familiar with Bloomberg's user culture. I don't know that I buy that it's going to be replaced with LLM chat prompts. But software agents are going to make faithfully reproducing those existing user interfaces much easier, so: half credit?
2. Half credit again on LLMs vaporizing the "business logic" moat, because the vertical-specific rules that justified the original software market are I think a lot harder to encode in Markdown than the 1 week they gave it, and also verification becomes a bear as more ground-truth business logic is replaced with nondeterministic AI output. There's a thing happening here for sure, I just don't buy it's as decisive as they say.
3. Public data access: I 100% buy this. If this was a real moat, it's dead.
4. Talent scarcity: same deal. Remember, we're talking about vertical software, where the underlying technical work is fairly repetitive and best-practices driven; it's the exact slice of software development work LLMs excel at.
5. Bundling (you get IB messaging along with your charting and your news service); maybe. This point feels tautological. Work out what LLMs do to each of the bundled experiences and there's your answer for how resilient that moat is.
6. Proprietary data: I think they're just dead on right here, and it does indeed seem to be a good time to be a company like Bloomberg?
7. Regs lock-in: half credit, because AI does make regs compliance a lot easier, and I think we're at the very early stages of seeing how.
8. Network effects seems like a repetition of "bundling" and if I have a qualm about this rubric it's that they made it look like an even 10, so they could have clean wins and losses.
9. Transaction embedding (ie, being a payment processor or a loan originator) also seems tautological; it's a moat, sure, but they're begging the question of whether AI enables people to stand up viable competitors.
10. I think "system of record" and "transaction embedding" are kind of the same moat.
I wish people would not blog on X (I will call it X when it's used as a crappy blog platform); these ASCII charts are awful. But that's neither here nor &c.
> Please don't use HN primarily for promotion. It's ok to post your own stuff part of the time, but the primary use of the site should be for curiosity.