AI vs. software companies

[Longnose]

Service Now posted strong earnings boosting confidence that AI is not killing Software.

[MungerWunger]

CSU barely pumped. SAD!

 

JK, hoping it goes to $ 1500 per share so I can full port it  

[KPO]

1 hour ago, MungerWunger said:

CSU barely pumped. SAD!

 

JK, hoping it goes to $ 1500 per share so I can full port it  

I bought on Wednesday and it’s already up $175. I wasn’t expecting that and was planning to buy more, but it definitely moved a bit. Fascinating company. I was very late to it, but better late than never! 

[Spekulatius]

Reduced some software exposure here. Y’all know that Claude 4.8 came just out destroying everything in its path.

[MungerWunger]

9 minutes ago, Spekulatius said:

Reduced some software exposure here. Y’all know that Claude 4.8 came just out destroying everything in its path.

Is this the first time where new model release didnt tank software? They're also releasing mythos in the next few weeks and that didn't cause a massive sell off lol

[Spekulatius]

1 minute ago, MungerWunger said:

Is this the first time where new model release didnt tank software? They're also releasing mythos in the next few weeks and that didn't cause a massive sell off lol

I feel like I just run out of a burning house with only my pants on fire.

Anyways, if software sells off again, I am going to increase my exposure again. The whole thing had been a humbling experience to say the least. I do love the volatility however.

 

Bought some ADSK this morning after it inexplicably sold off after earnings this morning but then software caught a bit.

[Cor]

1 hour ago, Spekulatius said:

I feel like I just run out of a burning house with only my pants on fire.

Anyways, if software sells off again, I am going to increase my exposure again. The whole thing had been a humbling experience to say the least. I do love the volatility however.

 

Bought some ADSK this morning after it inexplicably sold off after earnings this morning but then software caught a bit.

They announced a major acquisition to the tune of ~3.5B, largest in their history I believe. The sell off could have been a reaction to that? 

[formthirteen]

4 hours ago, MungerWunger said:

Is this the first time where new model release didnt tank software? They're also releasing mythos in the next few weeks and that didn't cause a massive sell off lol

 

Checks portfolio... My bet is on software going up from here, so I think we agree?!

[thowed]

10 hours ago, Longnose said:

Service Now posted strong earnings boosting confidence that AI is not killing Software.

 

Service Now has become a meme stock - Trump (reportedly?) has been buying, and so... well, y'know.

 

ADSK - yes, believe the acquisition was not good (overpriced?) so poor reflection on management.  Hopefully business continues well anyway - I am an interested party by proxy as an owner of a slug of RWWI.

 

Hoping software has bottomed now (as I have enough!) - Mythos, yadda yadda - well, I'm too old to have a clue (but who does really) and have my own biases - but feel like there have been credible recent arguments that AI will improve software companies, rather than disrupt them, and that there are aspects of the business (human communication & selling, trust etc.) which AI cannot do.

[Cor]

My personal opinion on software vs AI for what little it’s worth is this: AI may help smaller companies build software that matches what the big boys already have and maybe even better by adding novel features more quickly vs having to go through all the red tape at a large company. However, because SaaS customers are often enterprise and not B2C (and this is key), I can’t see a reality where customers stop doing business with say someone like Salesforce to switch to a half priced AI-native alternative (even if you assume that the switch is frictionless). 
 

In enterprise SaaS things can go wrong. When businesses depend on the tools for critical business processes (think ERPs,  inventory management systems, etc) they want to be dealing with a reputable SaaS vendor, not a startup. I am of the opinion that startups cannot support medium/large enterprise customers nearly as well and large companies can. Then come issues of cybersecurity, uptime/reliability…
 

At the end of the day, SaaS is not THAT expensive even at $1M/year for a business worth billions. Even taking that expense down to $100k/year shouldn’t make a material difference to quarterly earnings unless I’m really missing something here? 

 

My assumption here is that most large SaaS companies would have the bulk of their revenue coming from mid-market/enterprise accounts who would not trade cost savings for vendor reputation, supports, security and uptime. If the core base of customers is SMB, I would think differently about the risk to a SaaS company in this new environment. 

[NnnnotSoSmart]

Saaspocalypse? Or just software back to market multiples? Slide below.

 

Starting at 10:52 Gerstner: Most of these software names are trading at a higher multiple than NVDA.

 

NVDA is trading at 13 times earnings for a company that is growing at 70%, for the thing that is the most essential thing to AI. And yet arguably AI challenged software is at twice the multiple.   

 

[Spekulatius]

On 5/29/2026 at 9:48 PM, Cor said:

They announced a major acquisition to the tune of ~3.5B, largest in their history I believe. The sell off could have been a reaction to that? 

Yes, the fit seems quite good, but the valuation of the acquisition not so much. i am not sure its the reason for the decline because Mr Market changed his opinion a couple hours later. 

[Libs]

36:30 mark - Jordi Visser with a clip of an analyst endorsing SAAS. 

 

 

[hardcorevalue]

On 5/31/2026 at 2:07 PM, NnnnotSoSmart said:

Saaspocalypse? Or just software back to market multiples? Slide below.

 

Starting at 10:52 Gerstner: Most of these software names are trading at a higher multiple than NVDA.

 

NVDA is trading at 13 times earnings for a company that is growing at 70%, for the thing that is the most essential thing to AI. And yet arguably AI challenged software is at twice the multiple.   

 

 

NVDA is 23x jan 2027 earnings. 

17x jan 2028. 

no idea what he's talking about

[frommi]

Microsoft has introduced token based pricing at the start of june for copilot. This is the assumption that Gemini has given me: However, early data from the transition indicates that heavy-use environments can see cost multipliers ranging from 5x to 9x their previous baseline.

Even if this doesn't break AI adoption i can see it being a big brake for most companies. Uber already introduced caps of 1500$ per employee/month.

Summary: For the average, high-performing software engineer, this cap is intended to be a "guardrail" rather than a blocker—it allows for roughly 1–2 hours of intensive AI-agent usage every single workday. Anything beyond that will require you to seek manager approval or optimize your prompts to be more token-efficient.

Edited 15 hours ago by frommi

[Milu]

24 minutes ago, frommi said:

Uber already introduced caps of 1500$ per employee/month.

And people think that there won’t be a return on investment for these AI investments? 18k per year is a lot more than the $1200 or so per year for normal enterprise SAAS companies like Salesforce.

[frommi]

42 minutes ago, Milu said:

And people think that there won’t be a return on investment for these AI investments? 18k per year is a lot more than the $1200 or so per year for normal enterprise SAAS companies like Salesforce.

That 15k is for 1-2 hours per day. Hard to see more than 10-15% productivty gain with that timeframe. Cost per employee also goes up 10%, so you see productivity increase 0-5%. Hardly groundbreaking.

[Spekulatius]

3 hours ago, frommi said:

That 15k is for 1-2 hours per day. Hard to see more than 10-15% productivty gain with that timeframe. Cost per employee also goes up 10%, so you see productivity increase 0-5%. Hardly groundbreaking.

Depends on how much more work they get done in 1-2 hours. I think companies will find a way to use the cheaper models (opensource models, Deepseek) more and reserve the expense usage based models for the critical stuff that needs it. The Uber CEO said so much. There is already software around that automatically switches from one model to another based on presumed fit to save costs. AI is going to have a control layer on top to prevent misuse, save costs as well as for cybersecurity.

 

Edited 10 hours ago by Spekulatius

[Milu]

1 hour ago, frommi said:

That 15k is for 1-2 hours per day. Hard to see more than 10-15% productivty gain with that timeframe. Cost per employee also goes up 10%, so you see productivity increase 0-5%. Hardly groundbreaking.

Hard to say, I personally use Claude each day for my job in Salesforce development and it’s allowed me to complete tasks in 30 mins that would ordinarily have taken 4 to 8 hours. I suspect the same for other developer use cases. I think we will eventually get to a stage where each employee is given some sort of token budget each month to use as they please. So an employee cost will be some combo or salary plus token cost. For AI early adopters maybe the ratio is 80% salary 20% token, and for more conservative companies maybe 95% 5%. I expect the portion of token percentage to trend upwards over time.

[frommi]

We still work in agile teams to do our coding and since we use AI our productivity hasn't changed, we still do the same amount of story points per sprint. Maybe thats just us, or the productivity leads to more leisure time for the developers. My company just has higher spending because of AI. Thats maybe why i am biased.

[Milu]

28 minutes ago, frommi said:

We still work in agile teams to do our coding and since we use AI our productivity hasn't changed, we still do the same amount of story points per sprint. Maybe thats just us, or the productivity leads to more leisure time for the developers. My company just has higher spending because of AI. Thats maybe why i am biased.

Yes I think you are right, the development approach hasn’t really changed, so even though it takes much less time to do things, the estimates for the user stories and what has been sold to the client is based in the old world. So developers are just delivering a standard chunk of work faster and probably chiliing more with nothing to do. So yes while developer productivity has improved massively in the sense of delivering the same chunk of work faster it doesn’t yet show up in actual productivity numbers because the pre AI block of time estimated or sold for the work hasn’t factored in the change in speed of what can be delivered. 

[frommi]

11 minutes ago, Milu said:

So yes while developer productivity has improved massively in the sense of delivering the same chunk of work faster it doesn’t yet show up in actual productivity numbers because the pre AI block of time estimated or sold for the work hasn’t factored in the change in speed of what can be delivered. 

I am not really sold on that one either, because at the end of every sprint we typically still have stuff thats left over. If developers really are faster now, that should go away, but it doesn't. We still have the same amount of stuff that didnt get done in each sprint that was planned to be done. Sometimes i think that you just use that time now to write the prompts instead of the code. And because the amount of code is now higher it just feels that you gained speed, but its even worse because the quality of code is now lower.

[Spekulatius]

2 hours ago, Milu said:

Yes I think you are right, the development approach hasn’t really changed, so even though it takes much less time to do things, the estimates for the user stories and what has been sold to the client is based in the old world. So developers are just delivering a standard chunk of work faster and probably chiliing more with nothing to do. So yes while developer productivity has improved massively in the sense of delivering the same chunk of work faster it doesn’t yet show up in actual productivity numbers because the pre AI block of time estimated or sold for the work hasn’t factored in the change in speed of what can be delivered. 

Similar to how everything being online and easy to access  via PC or smartphone has made everyone more productive but also more distracted and the same tools make it easy to waste time.

Edited 6 hours ago by Spekulatius

[patience_and_focus]

2 hours ago, frommi said:

I am not really sold on that one either, because at the end of every sprint we typically still have stuff thats left over. If developers really are faster now, that should go away, but it doesn't. We still have the same amount of stuff that didnt get done in each sprint that was planned to be done. Sometimes i think that you just use that time now to write the prompts instead of the code. And because the amount of code is now higher it just feels that you gained speed, but its even worse because the quality of code is now lower.

Do you feel like the reason for that has more to do with things like cross functional decision making, waiting on feedback (internal or external) to plan next steps that then bleed over to the next sprint, infra blockers (simple things like granting iam permissions but requires signoff), etc?

 

The reason why I ask is because it points to the fact that considerable portion of software development (coding specifically) was done in parallel while many such non AI issues were sorted out. Now even with the coding time shrinking, the other parallel things / processes needed to for delivery are just as slow as before. So overall we are still bounded by that slowest and  parallel processes. And therefore productivity doesn't increase proportionally.

Edited 6 hours ago by patience_and_focus

[nsx5200]

This is the same problem that teams have when deciding to off-shore development: spend time to formalized requirements/prompts, or spend time to do the actual design/development, with the hidden "requirements" communicated via human interactions(meetings/chats/"culture"/philosophy).  With the off-shore teams, it's difficult to communicate culture and philosophy, so the end-product tend not to mesh as nicely.  AI is an extreme form of off-shoring, with even fewer channels for communicating "culture"/philosophy correctly.  I'm pretty sure this thread will explode if we start going down the rabbit holes of horror off-shore stories.

 

If vibe coding works as marketed, then product managers would be writing the requirements for AI to code, but it seems like the reality is that people w/ formal training in software engineering is still needed to write those requirements for AI coding right now.