AI vs. software companies

[Castanza]

1 hour ago, rogermunibond said:

Can AI/LLMs reduce the number of CAD seats?  Thinking ADSK etc.

 

1 hour ago, frommi said:

Of course, who needs CAD when the robot builds what you want from a prompt?

 

Web Development jobs and the TAM has done nothing but grow since the release of "vibe code" products from SquareSpace, Wix, etc. 

 

Most of these "can AI" or "will AI do this" questions are useless exercises. There is no point contemplating a pie in the sky question that ignores the nuance the real world demands. 

[sholland]

5 hours ago, rogermunibond said:

Can AI/LLMs reduce the number of CAD seats?  Thinking ADSK etc.

As a mechanical engineer who has used CAD for over 30 years, it is hard for me to fathom AI/LLMs reducing the number of CAD seats.  When I used to work in R&D I used to say if we knew what we were doing we wouldn’t call it research.  New designs are very much an iterative process with lots of testing.  Even after product launch it was a rare drawing that didn’t have any revisions. 

 

Perhaps the manufacturing engineers wouldn’t need as many seats as I could imagine AI figuring out how to manufacture the designs, but that seems like a marginal impact in the distant future. 

[gfp]

Yes, my friend who works at BSY - Bentley Systems (an ADSK competitor in some products) says their niches are generally not at risk.  Did mention a lot of layoffs at ADSK, which allowed BSY to pick up some new hires.

[Spekulatius]

2 hours ago, gfp said:

Yes, my friend who works at BSY - Bentley Systems (an ADSK competitor in some products) says their niches are generally not at risk.  Did mention a lot of layoffs at ADSK, which allowed BSY to pick up some new hires.

ADSK is starting to look interesting too. They have been running leaner , so even if the topline does not grow that much, the bottom line grows almost twice as fast. That’s something we are going to see with a lot of software companies , I think. There are some lower hanging fruits during customer support roles, management layers and probably ins software engineering as well.

[MungerWunger]

Akre fund added to NOW and CRM: https://www.akrefund.com/fund-summary/

[Eng12345]

Watching a few youtube videos - claude for excel is very impressive. 

 

I'm looking forward to playing around with it and seeing how well it works

 

[MungerWunger]

Akre letter: https://www.akrefund.com/documents/a-letter-to-the-investment-team-february-6-2026/

[Dalal.Holdings]

WIX not gonna survive AI IMO

 

 

 

A lot of these software businesses also have a lot of stock based comp so the actual earnings to shareholders are not very good

[Spekulatius]

This one ($HUBs) will survive, IMO.~20% revenue growth and they seem to aggressively using AI agents to make their product more replant and easier to use. it’s unlikely that customer with 1000 employees build their own software when there is an off the shelf solution.

 

Edited February 13 by Spekulatius

[MungerWunger]

https://cheekypint.substack.com/p/ben-thompson-from-stratechery-on

 

Ben Thompson from Stratechery on AI ads, the end of SaaS, and the future of media

Edited February 13 by MungerWunger

[UK]

2 hours ago, MungerWunger said:

https://cheekypint.substack.com/p/ben-thompson-from-stratechery-on

 

Ben Thompson from Stratechery on AI ads, the end of SaaS, and the future of media

Thanks!

[MungerWunger]

https://x.com/TechFundies/status/2022041255145083313

 

"Anthropic held a webinar for Claude for Excel. They didn’t do a real-time demo since, as anyone who has tried it knows, it takes a long time and doesn’t work very well for anything beyond the simplest of requests. 

 

They appear to be trying to get AI to work for knowledge work the same way it became useful for programmers. The web AI doesn’t work at all for one-shot, complex requests on files (which is most work people do…) so it’s smart to bring AI into the app (thanks Excel) and encourage people to try it for baby step tasks (format this section, add pivot table, etc.). 

Programmers mostly use it to work on blocks of functionality where AI does an increasingly good job. So the view seems to be let’s try that path for app usage and meet people where they are. 

 

2 customers joined the call. DE Shaw basically said they try everything AI in a research capacity in case it starts to work but for now they don’t really know what to do with it. 

 

RBC said AI helping them a lot across different areas (research, etc) but also hard to gauge ROI and the excel stuff is very early / makes lots of mistakes. He is paid to research this stuff but people who have business deliverables aren’t going to be as patient. 

 

One of the customers also mentioned all this experimentation is getting too expensive and hinted it will be an issue soon. 

The Anthropic marketing guy basically said look all this is early but we encourage you to not give up bc this isn’t a product yet. Please keep spending your money beta testing our stuff for us. There was nothing that would suggest MSFT is going to be any worse off from any of this. Quite the opposite actually."

[frommi]

We’re mostly hearing the 'blue sky' projections from AI companies and their CEOs, but the operational reality is far less polished. The moment you move from demos to deployment, the fundamental flaw of probabilistic modeling hits a wall. In creative fields, unpredictability is a feature; in mission-critical infrastructure, it’s a liability. No serious operator is going to deploy a system with a 1% 'catastrophic failure' rate. We’re essentially looking at a machine that blows its own engine every hundredth cycle—that’s not an asset, it’s a tail-risk nightmare.

Edited February 13 by frommi

[Castanza]

4 hours ago, frommi said:

We’re mostly hearing the 'blue sky' projections from AI companies and their CEOs, but the operational reality is far less polished. The moment you move from demos to deployment, the fundamental flaw of probabilistic modeling hits a wall. In creative fields, unpredictability is a feature; in mission-critical infrastructure, it’s a liability. No serious operator is going to deploy a system with a 1% 'catastrophic failure' rate. We’re essentially looking at a machine that blows its own engine every hundredth cycle—that’s not an asset, it’s a tail-risk nightmare.

Agree 100%
 

AI is probabilistic 

 

Traditional programming is deterministic 

 

An important distinction for people to remember when thinking about this topic. 

[Spekulatius]

LOL, so true.

 

[Valuebo]

5 minutes ago, Castanza said:

Agree 100%
 

AI is probabilistic 

 

Traditional programming is deterministic 

 

An important distinction for people to remember when thinking about this topic. 

 

Thank you for the common sense! Glad to see it still exists.

[Spekulatius]

23 minutes ago, Castanza said:

Agree 100%
 

AI is probabilistic 

 

Traditional programming is deterministic 

 

An important distinction for people to remember when thinking about this topic. 

Yes, and everything the AI “knows” is scraped from existing data and out into new context based on the prompts asked. AI cannot invent anything new. It cannot reason, although it seems like reasoning because the context is based on reasoning done before (from the data scraped).

 

So instead of just the sources that you have become available through search at your fingertips, you get now all the existing context available to you and at your fingertips through LLM.

[Hektor]

AI will impact many software companies. Here is a simple example.

 

The Value Line Investment Survey subscription costs about $50/month. A Claude.ai Pro subscription costs $17/month. Tell Claude.ai to create a report from the 10-K of any public company by feeding it the names of all the rows in the VL report, and you have it within minutes, if not seconds.

[frommi]

Of course some companies will be affected, but not all. And right now the market just throws everything into the same bucket. Where do you think Claude gets its financial information? 

[Hektor]

4 minutes ago, frommi said:

Where do you think Claude gets its financial information?

I assume its either the IR webpage or Edgar

[Castanza]

1 hour ago, Spekulatius said:

Yes, and everything the AI “knows” is scraped from existing data and out into new context based on the prompts asked. AI cannot invent anything new. It cannot reason, although it seems like reasoning because the context is based on reasoning done before (from the data scraped).

 

So instead of just the sources that you have become available through search at your fingertips, you get now all the existing context available to you and at your fingertips through LLM.

 

It's truly amazing....with the way the market is reacting both positively and negatively to AI; one would think P!=NP was solved...So far AI bridges the time gap to discovery utilizing known methods. This is a massive advantage that seems (or sounds) like magic to many. But Humans are still at the helm when it comes to chosen direction. Rules are meant to be broken no doubt (Turing test); however it seems most experts (not PR hacks) agree that there is a handful of significant breakthroughs needed in the field to bridge the divide between reality and theory. IF (big IF) that day ever comes, who knows what the world will look like....

 

The real question imo is how far can we get with the current approach? I think we will be surprised on both the upside and downside. 

 

[frommi]

44 minutes ago, Hektor said:

I assume its either the IR webpage or Edgar

In part yes, but its also connected to factset's and S&P database API for example.

[Spooky]

3 hours ago, Spekulatius said:

Yes, and everything the AI “knows” is scraped from existing data and out into new context based on the prompts asked. AI cannot invent anything new. It cannot reason, although it seems like reasoning because the context is based on reasoning done before (from the data scraped).

 

So instead of just the sources that you have become available through search at your fingertips, you get now all the existing context available to you and at your fingertips through LLM.

 

I'm not so sure about this. Look at Google's Alpha Go - it came up with a totally new strategy to win a game that has been played by humans for over 2,000 years.

[gfp]

16 minutes ago, Spooky said:

 

I'm not so sure about this. Look at Google's Alpha Go - it came up with a totally new strategy to win a game that has been played by humans for over 2,000 years.

 

... that occurred 10 YEARS AGO

 

 

[yesman182]

Anyone care to recommend at AI service? I have never paid for an AI service. I assume I should get with the times. Most helpful things for me would be reviewing 10ks and quarterly calls. As well as consolidating my notes.  Self employed so I don’t need it to work with any specific software.