After the SaaSacre
For most of the past two decades, SaaS was the best business model anyone had built. Then last week, $300 billion in market cap vanished across Figma, Salesforce, and most names in between. Evan Armstrong captured what the market decided. “Nah.”
The fear driving the selloff is that AI-generated code makes software worth less. Armstrong thinks the market is right about that, but wrong about what comes next. Three other analysts published this week and, without citing each other, landed on roughly the same conclusion.
QUARTERLY BRIEFING
Forward Outlook with Eric Rosen and Mike Schatzman
A quarterly economic briefing with Eric Rosen, Economic Advisor to Venture Forward Capital.
Monday, March 23, 2026 | 2:30 PM CT·
The Three Layers and Where Value Goes
Armstrong thinks the selloff is partially justified. A lot of these companies genuinely deserve to be valued less. But the market is punishing all software for what’s really happening to some of it. The SaaSacre, he writes, is “a culling of the herd, a separation. Software is splitting into three layers with fundamentally different economics. Two of those layers are getting commoditized. The third—a layer that barely existed before AI—is where the value migrates.”
Systems of record “didn’t die when we switched to cloud, but they stopped commanding premium growth multiples.” They’re databases. They have to exist so agents have something to reference. Per-seat tools built for individual humans, the project management software and analytics dashboards, are what Armstrong calls “the true SaaSacre zone.” Already 4% of GitHub commits are generated by Claude Code.
The third layer never existed as software before. Armstrong calls it the context layer, the institutional knowledge that tells agents what to do, in what order, and whether they’re allowed to do it. Before AI, it was “the email threads, wiki pages, Slack channels, onboarding docs, and tribal knowledge where organizational truth actually lived. It was never structured. Never searchable. Never maintained. It was just the ugly overhead that required companies to hire more people than they wanted.”
Armstrong thinks that ugly overhead is about to become the most valuable software in the stack. “Generating code is commoditized,” he writes. “Governing code in production is not.” A markdown file can describe your sales process. “It can’t encode that deals over $500K stall when legal reviews before procurement.” The context layer, he writes, “is what makes AI agents productive instead of just active.” And it compounds, because every workflow an agent executes feeds back into the layer, making the next execution smarter.
We wrote about this shift in December. Clayton Christensen had a name for this. The Law of Conservation of Attractive Profits. When one layer commoditizes, the adjacent layer de-commoditizes. IBM’s hardware commoditized and value migrated to Intel and Microsoft. Applications and databases are commoditizing now. Armstrong leaves it as an open question. Does the context layer get owned by whoever already holds the organizational knowledge, or whoever builds the most capable agents?
The Revenue That Looks Real and Isn’t
Before anyone can own the context layer, they have to survive long enough to build it. Augustin Sayer thinks most AI startups won’t.
Sayer offers an example. A cybersecurity company goes from $100K to $1.5M in a year. A 15x jump. Once upon a time, that was the signal for any serious Series A fund to lead an outsized round. Today it gets filed under good, not great. Meanwhile an AI app ramps from zero to millions in ARR in months and triggers a feeding frenzy.
Sayer has a term for what that revenue is actually made of. Rented demand. “A meaningful slice of AI startup revenue behaves less like traditional SaaS and more like rented demand: easy to start, easy to switch, easy to price-compress.”
Switching is easy because the cost to query a model at GPT-3.5 performance dropped from roughly $20 per million tokens to about $0.07 in under two years. When the floor drops that fast, any competitor can match you on price overnight. And the margins tell the rest of the story. Classic SaaS ran at 75 to 85% gross margins. Many AI-native companies operate closer to 25%. “Fast growth plus low gross margin is not the same asset as fast growth plus 75 to 90% gross margin.” Spikes are not moats.
The risk, Sayer writes, is that “we’re turning venture into a bad trade: paying peak prices for revenue that’s cheaper to replace every quarter, while starving the companies building the boring, embedded, compounding value that creates real customer lock-in.” “In 3 to 7 years, the scoreboard won’t reward who grew fastest from zero. It will reward who built the revenue that stayed.”
The Check Engine Light
Sayer is describing the problem from the investor side. From inside Google, Darren Mowry is seeing which startups are already showing symptoms.
Mowry runs Google’s global startup organization across Cloud, DeepMind, and Alphabet. Two common AI startup models, he says, have their “check engine light” on.
Not all LLM wrappers are created equal. Cursor and Harvey AI have built deep integration, proprietary domain data, performance that compounds with use. Most wrappers haven’t. Wrapping “very thin intellectual property around Gemini or GPT-5” signals you’re not differentiating yourself, Mowry says. “If you’re really just counting on the back-end model to do all the work and you’re almost white-labeling that model, the industry doesn’t have a lot of patience for that anymore.” What founders need, he said, are “deep, wide moats that are either horizontally differentiated or something really specific to a vertical market.”
The second type is the AI aggregator, startups that built a business letting companies access multiple AI models through one platform. The problem is that Microsoft and Amazon are now bundling the same thing into Azure and Bedrock as a standard feature. Mowry’s advice was simple. “Stay out of the aggregator business.”
The parallel he draws is the early days of AWS. Startups sprang up to resell cloud infrastructure with better tooling and billing. But when Amazon built its own enterprise tools and customers learned to manage cloud services directly, most of those resellers got squeezed out. The only ones that survived had added real services on top, like security, migration, or DevOps consulting.
What If the Bulls Are Right
Everything so far has been about which startups survive. Citrini Research and Alap Shah go further. They’re asking whether the entire economy does.
They published a piece written as if looking back from June 2028. “What if our AI bullishness continues to be right,” they write, “and what if that’s actually bearish?”
The smaller vendors go first. Citrini describes a Fortune 500 procurement manager negotiating a SaaS renewal. The salesperson ran the same playbook as last year: a 5% price increase, the standard “your team depends on us” pitch. The procurement manager told him he’d been in conversations with OpenAI about having their engineers use AI tools to replace the vendor entirely. They renewed at a 30% discount. That was a good outcome, Citrini writes. The Monday.coms, the Zapiers, the Asanas had it much worse.
But the pain doesn’t stop there. The big vendors thought they were safe. But companies like ServiceNow price by the seat, and AI was making their customers need fewer people. When customers cut 15% of their workforce, they cancel 15% of their subscriptions. The same layoffs boosting customer margins are gutting the vendor’s revenue. So the vendors cut their own staff and pour the savings into AI. “The company that sold workflow automation was being disrupted by better workflow automation, and its response was to cut headcount and use the savings to fund the very technology disrupting it.” What else were they supposed to do? As Citrini puts it, “Sit still and die slower?”
This is not the Kodak story. Kodak saw digital photography coming and ignored it. These companies saw AI coming and did exactly what they were supposed to do. They adopted it, restructured around it, cut costs to fund it. Every individual decision made sense. But when the entire sector makes the same rational decision at the same time, it stops being rational.
Citrini doesn’t stop at software. The same loop that guts SaaS revenue also guts the white-collar workforce that drives most of consumer spending. He models it all the way out to mortgage stress, consumer spending, and private credit defaults. “The system turned out to be one long daisy chain of correlated bets on white-collar productivity growth.” Software, Citrini writes, “was only the opening act.”
Along the way, it spread to every business whose value was navigating complexity that people found tedious. Travel booking, insurance renewals, routine legal work. “We had overestimated the value of ‘human relationships,’” Citrini writes. “Turns out that a lot of what people called relationships was simply friction with a friendly face.”
Citrini doesn’t call it a prediction. He calls it a scenario. But every number in it is sourced, every mechanism already in motion somewhere, and nobody in the piece has to do anything irrational for all of it to come true.
None of this has happened yet. The S&P is still near all-time highs. “The canary is still alive.”
Where We’re Looking
The Citrini scenario is worth taking seriously, but it’s not the whole picture. The companies at risk are the ones selling general productivity into white-collar workflows, the Asanas, the Mondays, the tools that got adopted because they were easy and will get cut because they’re replaceable. Applied AI focused on physical infrastructure, healthcare, or compliance doesn’t live in that daisy chain. Those workflows are too specific, too regulated, and too expensive to get wrong.
At VFC, we’ve been watching the shift from seat-based pricing to outcome-based models for exactly this reason. Seats get cut when headcount drops. Outcomes don’t. And Armstrong’s context layer isn’t just a thesis, it’s a moat. The companies encoding why a $500K deal stalls, or how a claims workflow actually moves through a hospital system, are building something that compounds with every interaction. That’s not rented demand. That’s the revenue that stays.
Venture Forward Capital is a venture fund specializing in Applied AI investments. We back entrepreneurs building the platforms and vertical applications that make AI work in the real world.