Does AI Threaten Vertical SaaS?

How Layer Commoditization Cycles create winners & losers

Today’s post is a collaboration with Luke Sophinos. He writes one of our favorite Substacks, Linear: A Vertical Software Newsletter, and is an Operating Partner at Atomic.

Abstract
+ We trace the origins of "AI will kill vertical software"
+ That thinking persists today despite weak logical backing 
+ The better question is: what SaaS is threatened by AI?
+ We find LLMs mirror historical "Layer Commoditization Cycles"
+ Drawing from this, we can get an idea of probable losers and winners

“AI is Killing Vertical Software”

We’re convinced few people ever actually believed this statement.

Which is interesting, because the perception that this is a held belief by many—or at least that it is an active debate, somewhere in VC and founder land—remains fairly common. The first person we can find that that said this earnestly and publicly was David Friedberg, on the All In podcast several years ago. As a successful founder¹ with deep scientific experience and passions, he was one of the more thoughtful commentators. So naturally, many took it seriously.

Unpacking the Prediction

In late 2023, the “besties” of All In submitted their predictions for worst performing asset in the coming year. We were piqued, of course, when Friedberg gave his answer: Vertical Software. Clearly, we weren’t the only ones. Andrew Oved at Reformation posted this writeup the following week, summarizing Friedberg’s rationale from the pod nicely:

“AI has enabled companies to build, with no-to-low code, their own internal software solutions at a much lower cost than the price of purchasing comparable software from a 3rd-party. As a result of this new ability to build internal tools more seamlessly, there will be increased churn and pricing compression resulting in deflationary SaaS pricing power.”

Friedberg also shared an anecdote to illustrate the point. An engineer at his company—presumably Ohalo—replicated a supply chain data management tool they had been paying >$50k annually for prior. That allowed them to churn off the SaaS product. He suggested they might open-source it or sell it for cheap.

Surely, he continued, talented devs like that one would find other such opportunities, to the chagrin of many Vertical Software vendors.

The Semantics Problem

As with most “annual predictions,” this vision was pretty far off the mark, or at least has been so far—since 2023, vertical platforms of various stripes have been taking off in ways few models have. Not only leaders of the Vertical AI boom (e.g. Abridge, Rilla, Harvey, etc.) that are by most measures still software companies, but also pure-play Vertical SaaS icons like ServiceTitan or Toast. But one could play Devil’s Advocate: perhaps David was right about the trend, just a few years early.²

Perhaps AI is quietly killing Vertical SaaS as we speak and soon all but the biggest, most horizontal software platforms will be rebuilt internally thanks to Loveable, Retool, Bubble, Replit, Cursor, Windsurf, Bolt.new, and whatever comes along next. Further, is this even a vertical-specific consideration, or will AI so automate and democratize software development that most SaaS today is just walking dead?

No doubt, Friedberg’s “ambitious engineers” can do much more today. Bringing on several engineers used to be be move #1 after raising first capital—since 2019, average fist hires are coming in ~70% later.³ But even armed with AI, no one is going to reproduce, say, ServiceTitan on a whim. Could they rebuild its very first incarnation, a desktop app with basic trade business management features, 10x faster today? Probably. But to replace ServiceTitan, that engineer would have to resign, raise capital, compete with alternatives, and build it into the platform it is today.

The relevant AI threat level today, then, has nothing to do with industry. It comes down to an individual product’s scale, complexity, and defensibility. Simple SaaS tools of all stripes are at risk. Just as startup wedge products, in the face of better teams and more funding, are too. We’d wager Friedberg had a definition of Vertical Software closer to “small and narrow” rather than “industry-specific”… a semantic issue more than a fundamental mis-judgement.

Layer Commoditization

So we agree “tools” are replaceable but true “platforms” are not. But where do we draw the line of tool vs. platform? Is AI is pushing that line as LLMs evolve, moving the bar to be a defensible platform ever-higher? Will it approach an asymptote—meaning there is a portion of SaaS that is clearly safe today—or is all SaaS facing slow but sure obsoletion? To answer this, we can look to how similar paradigm shifts have happened in the past.

The history of technology is defined by progressive cycles of “layer commoditization.” We’re all familiar with the canonical examples: electric looms threatening the textile industry, mechanized manufacturing in autos, tractors in farming, ATMs in banking, etc. We wrote recently about a double-commoditization, in which Salesforce pulled compute from on-prem to SaaS, followed by AWS pulling it to a new IaaS layer. Each cycle sees the following:

1. Innovation: Some core aspects of the process are automated, increasing speed and allowing workers to reallocate time, albeit with some adaptation.

2. Pushback: Reasonable fears around loss of control and obsoletion of the worker causes pushback, with some hailing the end of X (as we know it).

3. Commoditization: The process is democratized, barriers to entry fall, a new “layer” in the process is born… but ultimately the market thrives with newfound productivity.

LLMs aren’t the first time such a cycle has hit software development itself.

From FORTRAN to LLMs

In the early 1950s, Grace Hopper invented the first compiler, the grand daddy of dev tools. Compilers translated human-interpretable “languages” into machine code, which up until that point programmers had to write by hand (or in barely abstracted assembly language), dealing with all the nuances of that particular computer. Then, in 1954, John Backus of IBM and his team—amazingly, of fewer than 10 people—created FORTRAN (short for Formula Translator), the first (successful) high-level programming language.

Existing programmers, however, pushed back. As Backus put it, “the resistance of the priesthood was such that the whole thing was likely to be ignored.” Languages—skeptics said—were less elegant, less controllable, and (unsaid but obvious) they had the potential to make some jobs obsolete. Despite those fears, the productivity gains were undeniable, running the majority of IBM’s code within 5 years. From 1950 to 1970, the number of programmers employed in the US would grow 30-50x.

In the era of LLMs, a new cycle of layer commoditization is underway in software development. Satya Nadella, CEO of Microsoft, laid out the ultra-bull case in a recent episode of an All In spin-out, BG2.

The notion that [SaaS] business applications exist, that’s probably where they’ll all collapse, right in the Agent era… once the AI tier becomes the place where all the logic is, then people will start replacing the back ends, right?⁴

His vision was pretty hand-wavy but I’d agree with the bearishness around traditional SaaS UIs and input-heavy systems of record. Just like with FORTRAN, lower-level activities are getting commoditized for both users and developers, allowing us to focus time and energy on the higher-level architectures and workflows that matter. Moreover, new layers are developing: obviously in LLM infrastructure, but perhaps also in inter-agent communication.

AI-driven commoditization is a powerful sign of progress. As in past cycles, there will be inevitable losers. But amidst the chaos, new winners (and new layers to win) emerge.

So Who Loses and Who Wins?

We see three main categories of vulnerable SaaS…

1. Hyper-Narrow: Developers of extremely niche (in use case more than necessarily market size) tools that are easily replicable with AI support. These are the nearest-term, clearest losers.

2. Obsoletion by LLM: SaaS offering a service LLMs naturally subsume (e.g. Chegg, Grammarly, Dragon). That said, companies with the resources and willpower to re-invent themselves AI-first may thrive. Duolingo is talking this talk.⁵

3. Hesitant Incumbents: Legacy SaaS providers who simply don’t offer AI functionality at parity quickly enough, or hold on too tightly to outmoded cash-cow products. Incentives matter. We (at Euclid) envisioned these slow deaths in our essay on the development of systems of intelligence here.

If development-support AI continues to improve, however, will the line keep moving to the point where ever more established SaaS platforms are vulnerable? Is there a coming wave of “Windsurf churn” off known-name software?

A quick analogy may help illustrate why the progression of SaaS replacement is likely asymptotic rather than indefinite. Imagine a pair of super-sneakers that imbues basketball players with a lifetime of conditioning. Overnight, high-school bench warmers become NCAA-competitive. Even the best NBA pros improve. But the impact isn’t nearly as drastic for the latter—the Currys and LeBrons already have a lifetime of training, and much of their relative edge is driven by nuanced game-sense. The enhancing effect, in other words, is logarithmic. The higher-level you already are, the less it will help. The same is more-or-less true for software businesses… especially when you add that everyone gets the magic shoes.

AI may be forcing SaaS to evolve in new ways but predictions of its demise are either drastically overstated, or more likely, just the latest iteration of an age-old marketing tactic. As Benioff can attest, nothing catches eyeballs like being the first to foretell the death of something popular, when there’s a hint of truth to it. On the contrary, however, there’s a strong argument to be made that AI is a massive accelerant to SaaS platforms—and to Vertical SaaS in particular.

We (Nic and Luke both) believe that Vertical AI is a key paradigm shift in “software.” In the future, every job in every industry will have some form of AI co-pilot (in the general sense, not in the “co-pilot” business model sense) that helps them do their job better. In some cases, Vertical AI will be able to sufficiently do a specific job in a specific industry end-to-end.

In effect, companies of the future will be powered by both human and AI labor capital—even if agents aren’t thought of, budgeted for, or monetized the same as employees (or even “replacing labor” on a net basis). In many cases, however, AI will simply serve as a critical pillar of a platform that plays the same role, if in a more powerful way, to that of modern software.

So is SaaS walking dead? Certainly, some vulnerable solutions are. And the shape of our systems of record will change forever. But vertical technology isn’t any more effected than SaaS as a whole—in many ways, it’s more defensible. And to infer that CRMs, PMSs, or ERPs will fall into a black hole… well, we just don’t see it happening.

Thanks for reading Euclid Insights! Please find our additional sources here.⁶

Do you know a founder thinking through an idea in Vertical AI? We’d love to help. Please reach out via LinkedIn, email, or here on Substack in the comments below.

1. David was the founder of Climate Corp, which sold to Monsanto in 2013 for ~$1B.

2. The most painful kind of “right” for VCs.

3. Carta (2025). AI Fundraising: The New Wave of Startups. Carta Insights.

4. Gurley, Gerstner (2025). Satya Nadella. BG2 Podcast.

LinkedIn Post from Ed Sim with the relevant clip.

5. Thanks to Colin Gardiner for pointing it out.

6. Cassel (2022). How John Backus' Fortran Beat the Machine Code 'Priesthood'. The New Stack.

Chan (2025). Did Satya Nadella really say SaaS is DEAD?. Medium.

Colibri Digital (2025). SaaS is Dead: How Microsoft's CEO sees the Future of Business Software.

Haldar (2025). When Compilers Were the 'AI' That Scared Programmers. Vivek Haldar.

McKinsey (2023). The Economic Potential of Generative AI: The Next Productivity Frontier.

Norman (n.d.). John Backus and the Development of FORTRAN. History of Information.

NWHM (n.d.). Grace Hopper. National Women's History Museum.

Oved (2024). AI is Not the Death of Vertical SaaS. Reformation Partners.