AI vs. software companies

[Spekulatius]

@UK That’s a pretty good video presentationand summary. MSFT trades at about 10x revenue. Any reduction in profit margins will absolutely compress this valuation. Even if no reduction in profit margin occur, an infrastructure heavy business cant sustain 10x  revenue valuation because FCF will be lower than earnings.

Edited February 16 by Spekulatius

[dpetrescu]

I’ve been investing in Autodesk for almost a decade, it has been one of the easiest investments of my life, always a little expensive in share price but such a solid moat. My only two thoughts - really questions. Either way, looking a decade ahead this is no longer in the easy pile

 

1. In design, it is not about the product but about the interface. Every minute detail of a building design needs to be controlled and adjusted. Starting with an ai generated floor plan only makes it more difficult - building a design gradually from concept is more efficient. So my question is - how will AI impact the human -> visualization/documentation interface. I can imagine AI for example could take a rough hand sketch and convert it into a 3d documentation for development.
 

2. The moat for ADSK and other similar companies is the network of licenses. If a designer documents construction drawings in Autodesk Revit (this is BIM, CAD died a decade ago) all the engineers, consultants, contractors, subcontractors, product manufacturers, suppliers, and owners, and all public agencies also need those Autodesk BIM licenses. So question is - can AI somehow neutralize that platform and make the RVT extension usable (editable and readable) with generic tools?

3. I think AI will not reduce design, but it will increase amount of design. Instead of generic details and similar designs, it could pave the way for every most minute detail being customized and unique (form of every door handle, custom faucets, etc)

 

4. I can imagine a scenario in which designers are completely leapfrogged. If AI eventually is able to manage code and technical aspects automatically, it could open up design to the masses. Even as an architect, I like the idea of design being democratized.

 

5. So far there is zero impact in the AE & construction industry. But investing is a forward looking thing - Who knows how far into the future the moat will be tested. 

 

 

Edited February 15 by dpetrescu

[TwoCitiesCapital]

The below is what I wrote in the CSU thread: 

 

Long post. Probably not interesting for most. But outlines my experience with 4-hours of work using the Pro version of Claude. I'm not yet impressed and the TL;DR conclusion is :

 

Quote

I'm sure this will get better - but with the above being my experience, it's really hard to see business relying on this stuff and getting rid of third party support/coders/experts/etc. 

 

 

My experience: 

 

Quote

To provide some important context: 

 

I basically have no formal coding experience. I have taught myself VBA, SQL, and Microsoft Excel logic through everyday use and have regularly used the functions to create buttons at work that pull data, manipulate it, and create outputs that largely automate certain excel based tasks at work. I have also played around a little with python at home. I am not anywhere close to what a junior coder/developer would be, but am quite a bit more adept than most of the population who don't do it professionally. 

 

I spent 4-hours on Claude today to accomplish...not a lot. 

 

Michael Burry linked a tutorial to build a trade analyzer on his substack using Claude. Is supposed to use APIs to scrape my trade data from multiple brokerages, pull them into an excel file, and run calculations on P/L, correlations, risk, positioning exposure, return calculations, etc. etc. etc. 

 

I followed the script exactly. What I got was a file filled with formulas to reference the trade data, but the APIs to pull in the trade data weren't working. Claude said it had network or security issues and couldn't run the APIs itself, but could write my python scripts to run them locally. But that also didn't work. So I exported trade data manually from IB and Schwab to my computer and asked Claude to format them correctly. Claude then created a script for each to format them. Then a script to dump that data into the initial workbook. And even after all of that, the initial workbook only worked if I could go in a manually update the fields from closing price and currency conversion rates for EVERY day for the last 365 days b/c Claude isn't able to scrape the web for that info and built that as a dependency for the formulas. So a request for a single simple script turned into multiple manual downloads, multiple scripts, and ultimately a useless tool because it still requires an insane manual input labor. 

 

The second thing I tried was adjusting python script I already downloaded from another source. The script will pull the options data for a certain ticker to see if they meet certain criterium and gives me a 'yes' or 'no' response in a pop-up window. The code is for a strategy of trading options around earnings. The frustrating thing is that this code can only be run for a single ticker at a time and there are several companies reporting on any given day.

 

So I asked Claude to create me an excel workbook with all 500+ of the S&P 500 companies, list their earnings dates, group them in such a fashion as those announcing after market close are grouped with the pre-market open announcements of the following day, and amend my python code so I could run the screen for all of the tickers in a particular grouping at one time with the output in the excel doc. Multiple attempts failed. First it couldn't scrape the 500 tickers for the S&P 500 from Wikipedia, so I agreed to let it hard-code them (which will result in manual edits required later). Secondly, it failed to successfully scrape web for announced earnings dates even though there are plenty of websites where this is available. Because we never got the earnings date data, I couldn't tell if it would successfully group them in the fashion requested (which is a bit more complicated than just looking at the date).

 

Ultimately, what I landed on is simply having it amend the python code to run for multiple tickers at a time. I'll have to manually look up which tickers are announcing on which day and manually enter the tickers myself and run it for each manual grouping. Basically, the productivity gain is that I click a button once to run for 10+ tickers instead of 10+ times, but still have to do all of the labor intensive work myself...

 

 

I'm sure this will get better - but with the above being my experience, it's really hard to see business relying on this stuff and getting rid of third party support/coders/experts/etc. 

 

Edited February 15 by TwoCitiesCapital

[Thrifty3000]

On 2/14/2026 at 7:48 AM, Spekulatius said:

Well, he works for Microsoft and they do not need raise money.

 

If these guys are remotely right, we need to buy puts on pretty much everything, especially financials. When half the people go in default, pretty much any financial stocks should crash and would be a zero.

 

Cash would be relatively safe in such a deflationary scenario, at least at first.

Gov will have to print trillions at first to stop the panic and trillions later to service debt

[Thrifty3000]

On 2/13/2026 at 9:53 PM, Spekulatius said:

I would like to take the other side of this bet.  
 

Also, if I were interviewer, I would have asked the him question how Microsoft HR is dealing with the problem. Are they starting the layoffs now or do they intend to do it all at once in 18 months? Is he going to be the last one to turn off the lights?

Until everyone on earth has everything they want Instantly there’s still opportunity to deploy capital, do work and provide value. (Unless some non-human species thinks energy used by humans is energy wasted. That would suck.)

[Buckeye]

Fund Beating 99% of Peers Sees Few Software Firms Surviving AI
 

https://finance.yahoo.com/news/fund-beating-99-peers-sees-090000453.html
 

 

[rogermunibond]

thanks @dpetrescu excellent insights on BIM.  are there possibilities where AI/LLM makes BIM more productive, such as building in construction/fire code limitations in the design process, or perhaps BIM already does this.

 

anyone have insight on the PD&M (product design & manufacture) business at ADSK? 

[Spekulatius]

2 hours ago, Buckeye said:

Fund Beating 99% of Peers Sees Few Software Firms Surviving AI
 

https://finance.yahoo.com/news/fund-beating-99-peers-sees-090000453.html
 

 

They may have left the software party a bit late. In their last 13F, their portfolio was chuck full with software stocks and now it seems they are aping into semis like everyone else. Will be interesting.

[AzCactus]

3 hours ago, Buckeye said:

Fund Beating 99% of Peers Sees Few Software Firms Surviving AI
 

https://finance.yahoo.com/news/fund-beating-99-peers-sees-090000453.html
 

 

If I'm not mistaken beating 99% was over a one year period-just saying.  

[tnathan]

On 2/12/2026 at 6:08 PM, Dalal.Holdings said:

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

The interesting thing about a company like Wix though is this -- for it to die it will require people to want to take their existing hosted website and move it to another platform / rebuild it. I'm not sure how many people are going to actually do that given the low monthly cost.

[Spekulatius]

15 minutes ago, tnathan said:

The interesting thing about a company like Wix though is this -- for it to die it will require people to want to take their existing hosted website and move it to another platform / rebuild it. I'm not sure how many people are going to actually do that given the low monthly cost.

Some will every month anyways through natural attrition. Many Wix websites are so simple that it’s easy to rebuild it and improve  If you start new, it’s unlikely someone will use Wix in the future. So you could easily see a 10% yearly shrinkage here.

[kab60]

On 2/16/2026 at 5:10 PM, AzCactus said:

If I'm not mistaken beating 99% was over a one year period-just saying.  

And apparently benchmarking themselves against Dow Tech instead of Nasdaq... Seems like just another grifter

[frommi]

This is probably the best description of what happens in software development right now:

 

[dcollon]

Really interesting.  Thanks for sharing the video 

[UK]

48 minutes ago, frommi said:

This is probably the best description of what happens in software development right now:

 

 

Gemini:

This video by Nate B Jones discusses the radical transformation of software engineering in 2026, driven by "dark factories" and AI agents that are hollowing out traditional career paths.
The Five Levels of "Vibe Coding"
The video introduces a framework by Dan Shapiro (CEO of Glowforge) to categorize AI integration in coding:
Level 0: Spicy Autocomplete – Basic code suggestions (e.g., original GitHub Copilot) [02:44].
Level 1: Coding Intern – Human gives a discrete task; AI writes the function; human reviews every line [03:02].
Level 2: Junior Developer – AI handles multi-file changes and navigates codebases, but humans still read all the code [03:30].
Level 3: Developer as Manager – The "flip" point; AI does the implementation and submits PRs while the human directs and approves at the feature level [03:56].
Level 4: Product Manager – Human writes specs and checks if tests pass; the code itself is a "black box" [04:41].
Level 5: The Dark Factory – Fully autonomous. Specification goes in, working software comes out. No humans write or review the code [05:14].
Key Concepts & Case Studies
The "J-Curve" Productivity Dip: Most developers actually get 19% slower when first using AI because they are "running a new engine on an old transmission"—trying to fit AI into legacy human workflows [14:44].
StrongDM’s Software Factory: A three-person team running a "Level 5" operation. They use "scenarios" (external behavioral specs) rather than traditional tests to prevent AI from "teaching to the test" or gaming the system [07:33].
AI Building AI: 90% of Anthropic’s Claude Code was written by Claude itself [00:58]. Similarly, OpenAI’s Codex 5.3 was instrumental in its own creation, resulting in a 25% speed improvement [11:44].
The Collapse of the Junior Dev Pipeline
The most striking data point is the decline in entry-level roles:
Junior job postings have dropped 67% in the US and over 50% in the UK [27:08].
The "Hollow Middle": Because AI now handles the simple bugs and features that juniors used to learn on, the "apprenticeship model" of software engineering is breaking [27:54].
The New Bar: A junior in 2026 needs the system-design skills of a 2020 mid-level engineer because implementation is now a commodity [28:40].
The Shift in Skills
The bottleneck has moved from implementation speed to specification quality [22:02].
Generalists over Specialists: Value now lies in understanding the "why" and "for whom" rather than being an expert in a specific narrow tech stack [30:04].
Organizational Deletion: Traditional structures like stand-ups, sprint planning, and QA teams are becoming "friction" rather than "coordination" in agent-led environments [19:21].
Conclusion: While the cost of software production is dropping by an order of magnitude—potentially exploding the total market for software—the "camouflage" of implementation complexity is gone. Only those with deep judgment and systems thinking will remain indispensable [41:45].

[dpetrescu]

On 2/16/2026 at 7:22 AM, rogermunibond said:

thanks @dpetrescu excellent insights on BIM.  are there possibilities where AI/LLM makes BIM more productive, such as building in construction/fire code limitations in the design process, or perhaps BIM already does this.

 

anyone have insight on the PD&M (product design & manufacture) business at ADSK? 

 

Autodesk already has chatbox AI assist. I imagine this will develop exponentially to be extremely functional. For now it feels like the teacher following 3rd grade students’ literal instructions on how to make a peanut butter sandwich.

 

what is interesting for the near term next few years is that all the huge AI build will require a lot of designers and engineers and builders and owners to buy BIM seats!

 

servers and energy generatin, utility operations centers and infrastructure is in a boom right now 

 

 

 

[Longnose]

This was really good. 

 

I agree, I would put myself into level 3 with some of the projects ive done so far.  Its gonna be a really interesting shift. 

I like that he acknowledges how challenging it will be for larger orgs to start leveraging and adopting this transition.  SAAS is not dead far from it.  But this shift will change the way that coding is done for these orgs. And it will take  substantial amount of time and money to make the changes required. 

 

But new smaller nimble players will start building some crazy shit in the next few years in the 4 and 5 levels and it will be cool to see what they build. But legacy software and systems will not be replaced overnight by AI coding Agents. 

 

 

7 hours ago, frommi said:

This is probably the best description of what happens in software development right now:

 

 

[MungerWunger]

 

[frommi]

There maybe is some truth and Anti-Hype in this video. And its funny

[roundball100]

On 2/18/2026 at 11:10 AM, frommi said:

This is probably the best description of what happens in software development right now:

 

A very persuasive speaker ... I wish I had a crystal ball to know how much of what he says is correct.  He appears to be incredibly knowledgeable about software engineering.  Assuming what he says is correct, then vertical-solution software companies (who know exactly what their present systems do, and precisely what their customers need - though this typically has never been written down in product specs, but rather evolved over time) are in much better shape than other software companies.  Because what those vertical-solution software companies know is what would be necessary to write the detailed, precise specifications needed for anything approaching Nate Jones' (this speaker's) Level 5.

[Milu]

Wow the bloodbath keeps going in SAAS stocks. They really are being seen as roadkill these days. Salesforce, Workday, Service Now, Hubspot, Atlassian, Adobe, Factset, Intuit. 

 

I'm almost tempted to just close my eyes and buy a basket of them, but probably just going to watch the fun from the sidelines. 

[jfan]

In my day job, I have a number of paper-based forms associated with quasi-governmental regulated entities, that require photocopying or faxing, coupled with logging into these entity portals, to manual do data entry in order to complete them. I'm not these entities primary client but a necessary cost center whose user experience and time is not considered something valuable. So this automating this with a better user interface is not really on anyone's priority list. 

 

So with no coding experience, I signed up for Claude, to build out an interface that solves this minute problem that others also deal with on a day to day basis. It is remarkable what it has built, but there certainly is a learning curve, requirement to download other pieces of software (eg database management, etc), in order to make this feasible. It's been about 5 days, and Claude has create for me a web login with 2-factor authentication, ability to adjust my profile/password, a clickable user interface. Hitting some snags getting it to print onto the designated fillable pdf. 

 

Anyways, the point(s) of my story is that AI coding has really unlocked people's ability to address various problems that could have a software solution, where previously it was impossible. I can only imagine the productivity gains software engineers might capture with these tools today. That said, the fear that software engineers will be eliminate is likely overblown. With my experience, it would be hard to imagine that my vibe coding product would be enterprise ready without any software engineering review and ongoing support. Furthermore, for truly mission critical workflows, I would hardly trust a non-technical person to vibe code and replace such legacy software. Finally, despite this simple problem, I'm still hacking away at it for 5 days without a functional product. I would imagine that the average person (non-software engineer) would have little to no interest in spending their time doing this (eg my adult relatives/in-laws, all were yawning at the dinner table as I blabbed on about my little side project).

 

With the problem surface area being pretty much infinite, I'm leaning towards that fact that AI will unlock some productivity (eg more prototyping, faster product releases) and the key skills to develop in this brave new world will be creativity, ability to empathize and understand other people's problems, critical thinking skills and continuous learning habits. 

[rogermunibond]

I have no dog in the SaaS vs AI debate as I never bought any SaaS stocks over the last 15 years.  Missed out on the way up for sure but there were a lot of things that never seemed right with the software stock uprating.

 

I keep wondering when the market figures out what these companies should be valued at, but I don't think it will be quick.

 

There are a couple things that make me think this is just early days. First, billions in PE-owned companies in PE funds need to shake out. Multiples are lower by probably 50-60%, debt was high for these companies 8-10x levered, and that all needs to shake out. Second, no one is taking any of the public SaaS companies private (with a few exceptions maybe) because most are not GAAP profitable. Growth investors are gonna get tired, and value investors aren't touching these names yet (with a few exceptions). It wouldn't surprise me if over the next few quarters SaaS company managers aren't looking at ways to get their costs in check, get GAAP profitable, and maintain their 30-40% revenue growth rate. They've all be called out and need to make their companies profitable.

[dpetrescu]

2 hours ago, rogermunibond said:

I have no dog in the SaaS vs AI debate as I never bought any SaaS stocks over the last 15 years.  Missed out on the way up for sure but there were a lot of things that never seemed right with the software stock uprating.

 

I keep wondering when the market figures out what these companies should be valued at, but I don't think it will be quick.

 

There are a couple things that make me think this is just early days. First, billions in PE-owned companies in PE funds need to shake out. Multiples are lower by probably 50-60%, debt was high for these companies 8-10x levered, and that all needs to shake out. Second, no one is taking any of the public SaaS companies private (with a few exceptions maybe) because most are not GAAP profitable. Growth investors are gonna get tired, and value investors aren't touching these names yet (with a few exceptions). It wouldn't surprise me if over the next few quarters SaaS company managers aren't looking at ways to get their costs in check, get GAAP profitable, and maintain their 30-40% revenue growth rate. They've all be called out and need to make their companies profitable.

 

Interesting question about market valuing SAAS. How can you value a company today that couple have a few years of growth and then is at Rick of falling off a cliff? There is a real risk a lot of SAAS companies with moats could be leapfrogged completely

[Spekulatius]

3 hours ago, dpetrescu said:

 

Interesting question about market valuing SAAS. How can you value a company today that couple have a few years of growth and then is at Rick of falling off a cliff? There is a real risk a lot of SAAS companies with moats could be leapfrogged completely

This is true for everyone, not just SaaS though. What if semiconductor chips are leapfrogged? Some of the same people  these claims have to problems paying up for memory business that for sure a re commodity like and where prices have trippled the last 6 month, but have a long long history of being on a deflationary trajectory. Or look at CAT trading at 5 x revenue because they sell some generators to AI datacenters as a filler for more durable solutions.

 

I don’t think the bubble is in software it is elsewhere.