Why SaaS Won't Come Back Yet

Look at the line the non-GAAP number is built to hide. GAAP EPS was $0.29 against the $0.90 non-GAAP figure everyone is quoting tonight. The gap between those two numbers is, by the company's own accounting, driven overwhelmingly by stock based compensation, which ran close to $1 billion for the quarter. ServiceNow's own management has publicly committed to getting SBC under 10% of revenue by 2029, which is itself an admission of roughly where it sits today, well above that. A company whose agents were genuinely replacing headcount should be able to show that showing up as a shrinking non cash comp line, not a target four years out. Nobody in tonight's earnings call is going to lead with the GAAP number. That's exactly why it's the one worth reading.

ServiceNow isn't a referendum on one company's growth rate. It's the closest thing the market has to a public test of the thesis that is currently keeping every SaaS multiple on the floor. AI agents are supposed to be replacing the humans a subscription used to require. If that's true, we should be able to see it. Not in a keynote. In a 10-Q.

Tonight's print didn't show that. Revenue accelerated. The stock rallied. The SBC gap didn't close. Not at ServiceNow, not at Salesforce, not in my own company.

The founder disclosure

I co-founded a SaaS company. We use Claude Code, agent harnesses, context engineering, prompt engineering, all of it, every week, since the tools existed. I want the 10x story to be true more than most people arguing about it online.

Here's what I've actually seen. A team of ten doing the work of eleven. Sometimes twelve. Not a team of ten doing the work of a hundred. That gap, between the story everyone is telling and what shows up in an actual P&L, is the whole article.

The selloff is real, the fundamentals aren't the problem

Software stocks lost more than $1 trillion in value in 2026 while the S&P 500 sat flat. The software sector ETF is down roughly 20%. $CRM and $ADBE are trading near a 50% discount to where they sat in early 2025.

Here's the part that should bother anyone shorting the category on fundamentals alone. SaaS revenue actually grew 17% year over year in Q1 2026, the fastest pace in three years, while large cap SaaS EV to sales multiples fell to 9.5x, a full standard deviation below the historical average. Growth held. The multiple didn't. That's not a fundamentals story. That's a narrative repricing three assumptions that used to be free money for this sector.

Oliver Wyman laid the three out cleanly. Software used to be protected because it was hard to build. Now agentic coding tools mean competitors, startups, or the customer itself can build it cheap and fast. Seat expansion used to be the monetization engine. Now agents do the work seats used to justify, so seat count shrinks. Features used to be the moat. Now agentic development commoditizes features faster than a sales team can defend them in a deal.

All three are bets on the future, not measurements of the present. That's the gap I want to close.

ServiceNow is the test case, not the exception

Tonight's number that matters most isn't revenue, it's this: ServiceNow's AI business crossed $1 billion in annual contract value this quarter, on pace for the $1.5 billion target McDermott set last quarter. cRPO hit $13.2 billion, up 21% year over year. That's real traction, and it's the single clearest AI monetization data point any SaaS company has produced this year.

It's also still layered on top of the per seat model the whole thesis says is under threat. ServiceNow just signed OpenAI as a preferred intelligence partner across the 80 billion workflows a year it runs for enterprises, alongside expanding partnerships with Amazon, Microsoft, Alphabet, and Anthropic, and OpenAI's own enterprise product line is now competing directly for the same IT service management budget. Subscription gross margin had already slipped from 84.5% to 82.5% heading into this quarter, on AI infrastructure spend, with management guiding FY26 to roughly 82%, so the AI revenue is arriving alongside AI cost, not instead of it. Nobody has proven the seats disappear yet. What's been proven tonight is that a $1 billion AI line can grow next to a cost structure that's getting more AI-expensive, not less, and that's before stock based compensation, still running close to $1 billion a quarter with management's own target to get it under 10% of revenue still four years out, is even part of the conversation.

Compare the two yourself instead of taking my word or the stock's word for it: ServiceNow vs Salesforce vs Twilio vs Adobe, side by side →

Where the evidence actually is, and isn't

Engineering is the function with the deepest AI adoption of any team in the business, and it's the best natural experiment we have. METR ran the closest thing to a gold standard trial: 16 experienced open source developers, 246 real tasks. The developers estimated AI would make them 24% faster. The controlled measurement showed them 19% slower, while they still believed they'd been faster. A separate longitudinal study across 400 companies found AI tool usage up 65% over 15 months, while actual pull request throughput rose only 7.76%. Researchers at MIT found measurably reduced neural connectivity in memory and creativity networks among heavy LLM users, and participants who struggled to recall what they'd just written.

I'm not citing this to say the tools are useless. I use them daily and they help. I'm citing it because engineering is the one function where we have controlled data instead of vendor claims, and even there, the honest number is single digit percentage gains, not a multiple.

Sales and marketing tell a similar story once you strip out the vendor surveys. Reps report AI cutting prospect research time by roughly a third and email drafting by a similar margin. That's useful. It is not a marketing team going from a hundred people to five while tripling conversion, which is the specific claim I keep hearing described secondhand and have yet to see one public company actually report with numbers attached.

Check this across the sector rather than trusting either the bulls or the bears: Enterprise software: retention, FCF margin, and headcount growth →

What would actually change my mind

Not another model release. Not another demo. A specific, boring, falsifiable thing: a public company reporting sustained double digit headcount reduction, held for more than one quarter, with flat or growing revenue and expanding margin, and management attributing it explicitly to agents in the filing, not the keynote.

Jason Lemkin's read on this cycle is the one I keep coming back to. Public SaaS growth rates have declined every quarter since the 2021 peak, and this year the market simply stopped pretending otherwise. AI is the accelerant. Slowing organic growth was the kindling that was already there. Dan Ives is on the other side of the same trade, calling the selloff a generational buying opportunity on the theory the market is pricing in disruption faster than it will actually arrive. Both of them are describing the same missing data point. Nobody has it yet.

The honest ending

Until a company actually shows the math, in a filing, not a slide, SaaS stays cheap and the AI infrastructure names keep absorbing the multiple. I don't know which quarter that changes. I own the same risk in my own company that I'm describing in this article, which is probably the most honest disclosure I can give you.

Go look at the actual numbers instead of the narrative: AI infrastructure: revenue per employee growth →