GM. This is Milk Road PRO, separating the software AI eats from the software it feeds.
Today's edition features our latest PRO report on the "SaaSpocalypse," the $2T wipeout that left software stocks priced like a dying industry. You'll get the opening below, with the rest on the site.
Here's what we've got for you today:
- 💀 The bear case for software: AI agents helped wipe roughly $2T off the sector, Salesforce fell about 30%, and software now trades below the average S&P 500 earnings multiple for the first time on record.
- 🔧 Our core thesis: AI kills software that just helps humans use tools, and feeds software that gets the work done. The market is pricing both for the same funeral.
- 🏗️ Three names that test it: ServiceNow, Salesforce, and Toast, each a different version of the survival argument.
Looking to diversify beyond stocks and crypto? Masterworks lets you invest in shares of blue-chip art like Picasso and Basquiat.
Prices as of 2:00 p.m. ET. Powered by CoinGecko.
THESE 3 STOCKS SURVIVED THE “SAASPOCALYPSE”
Yes, this is a PRO report about stocks, not crypto. Capital and attention have moved into AI, and the biggest mispricing we've found right now lives there, not in our dear digital tokens.
Wall Street has a new word for this gap in the market: the “SaaSpocalypse”.
In early 2026, the market erased nearly $2T of value from software stocks.
Salesforce fell about 30%. Workday 33%. The main software index lost more than 20% in a single quarter. And for the first time on record, software now trades at a lower earnings multiple than the average S&P 500 company. The market's golden child of two decades is priced like a dying industry.
The reason? AI agents: programs that do the work themselves. They resolve the ticket, update the record, draft the contract, and reconcile the invoice. If an agent can do the job, why pay for software that just helps a human do it?
I think that verdict is half right. And the half that's wrong is where the opportunity is.
My thesis in one sentence: AI kills software that helps humans use tools, and feeds software that gets work done. The market is pricing both for the same funeral. That's the mistake.
Below: the bear case (it's serious), the cracks in it (they're real), my five-pillar bull case, three companies to test it, and the exact numbers that would change my mind.
And if you want to skip all that and just see what I actually bought, get into Milk Road PRO for just a buck. You’ll see my full portfolio, my watchlist, and get to ask me anything you want in Discord.
THE BEAR CASE DESERVES RESPECT
The smartest version of the “SaaS is dead” argument goes like this.
1. Most software is a fancy filing cabinet. Microsoft CEO Satya Nadella said it best: most business apps are basically databases with some rules on top (you create records, read them, update them, delete them). If an AI agent can talk to the database directly, the screens, menus, and dashboards in between lose their reason to exist. The business logic moves into the agent.
2. The pricing model is built on a number that AI can shrink. Most software charges “per seat”: per human user, per month. If one employee plus AI does the work of three, the vendor sells two fewer seats. Research firm IDC predicts that by 2028, pure per-seat pricing will be mostly obsolete. This isn't theoretical anymore: Atlassian (SaaS company) reportedly logged its first-ever decline in enterprise seat counts in March 2026, and Workday (a company that sells workforce software) cut 8.5% of its own workforce.
3. New companies now grow at a speed that eats the incumbents. The classic comfort for incumbents was “startups can't reach enterprise customers fast enough.” Then Cursor went from zero to $2B in annual revenue in roughly three years. That is the fastest ramp in business software history, with most of the Fortune 1000 already using it. AI-native products spread like consumer apps, not like enterprise contracts.
4. The moat itself might be AI-soluble. Why do companies stay on old software? Because switching is painful: migrating data, rebuilding integrations, retraining staff. But AI is very good at exactly that kind of grunt work. If a 12-month migration becomes a three-month one, the pain that protected incumbents shrinks with it.
5. The model makers are climbing the stack. The selloff's biggest leg down came when Anthropic launched Claude Cowork, an AI desktop worker that drafts documents, builds spreadsheets, and runs multi-step workflows. OpenAI, Google, and Microsoft are shipping similar agents. If they deliver reliable agents with memory, permissions, and integrations, they don't need to build a CRM. They just absorb the long tail of workflow tools, one task at a time.
Put together, that's a real threat: AI attacks software wherever the product is just a human-facing screen for information work. I agree with this completely…but I also think it describes only a slice of the market, not the whole thing.
CRACKS IN THE FUNERAL STORY
Four things don't fit the “it's all over” narrative.
1. The numbers haven't cracked. Deutsche Bank made a blunt observation: it doesn't know of a single software company that expects AI to hurt its revenue this year. Q1 2026 earnings backed that up. Of the 16 software companies one Morningstar analyst covers, 14 beat on both the top and bottom lines, with guidance either in line or raised for the year and margins still expanding.
In the real software busts (2001, 2008, 2022), earnings collapsed alongside the stocks. This time the stocks fell and the earnings kept climbing. That gap has to close one way or the other.
2. The agents aren't ready. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027: costs balloon, the business value is fuzzy, and the risk controls aren't there. It estimates that of the thousands of vendors selling “AI agents,” only about 130 are the real thing; the rest are doing what Gartner calls “agent washing”: slapping the agent label on old chatbots. Today's models, in Gartner's words, lack the maturity to autonomously achieve complex business goals over time. Translation: the thing that's supposed to kill SaaS mostly doesn't work unsupervised yet.
3. The poster child changed its mind. Every "SaaS is dead" thread eventually cites Klarna (disclosure: I hold it personally), the fintech that cut hundreds of software tools (including Salesforce and Workday) and laid off 1,200 employees after building an AI customer-service system it claimed could do the work of 700 people. Less famous is what happened next: quality slipped, Klarna started rehiring humans, and the CEO publicly said he doesn't think this is the end of Salesforce. Reporters also found that Klarna didn't replace its software with AI: it mostly consolidated onto other software. The most-cited proof that AI replaces SaaS ended up proving something subtler: AI can do more than skeptics admit, and less than the headlines claim.
4. Nobody is actually leaving. The Q1 numbers show no churn wave, and not one major vendor has reported customers walking away for home-built AI. Even Nvidia's Jensen Huang (a man with every incentive to hype AI) called the "AI replaces software" theory illogical.
So the market is pricing a takeover by a workforce of agents that currently can't be left alone with the keys. That doesn't mean the bears are wrong forever. It means they're early. In markets, being too early can sometimes be the same as being wrong, if you’re on a shorter time frame.
LOOKING FOR BIGGER RETURNS?
Stocks are at their most expensive in 140 years.
So where do investors look when traditional markets feel expensive?
One option: blue-chip art.
Masterworks lets you invest in shares of blue-chip art like Picasso and Basquiat.
Why investors are paying attention:
- Returns 16.5%-17.8% on works held longer than a year
- Near-zero correlation with the S&P 500 since 1995*
- $1.3B deployed across 530+ artworks
If you’re looking to diversify beyond stocks, go check out Masterworks.
Click here to skip the waitlist and join
*According to Masterworks data. Investing involves risk. Past performance not indicative of future returns. See Reg A disclosures at masterworks.com/cd.
MY THESIS: PAY FOR WORK DONE
Companies poured an estimated $30-40B into generative AI, and one widely cited MIT study found that 95% of corporate AI pilots produced no measurable return.
Fair warning: the study is small and contested, but Gartner's project-cancellation forecast and every CFO we researched points in the same direction.
How can companies use this much AI and get this little out of it? Because there's a gap between what AI produces and what a business needs.
A model produces words, code, and suggestions. A business needs the invoice reconciled, the ticket closed, the shift scheduled, with a record of who approved it, rules about who's allowed to do it, and someone to call when it goes wrong.
Crossing that gap is called the translation layer. It's the scarce thing now. Raw intelligence is cheap…whereas trusted, verifiable, and complete work is not.
Here's what each layer adds, in plain terms:
Source: Milk Road
That right-hand column is where strong software already lives. The best software companies are systems of work: they encode how a process runs, who can do what, what gets recorded, and what gets escalated.
AI raises the value of that layer, because automated work needs a safe place to happen.
THE BULL CASE, IN FIVE PILLARS
1. Software has always been a replacement for labor. Payroll software replaced payroll clerks. CRM replaced pipeline spreadsheets. AI just extends that from paperwork to brainwork: writing, analysis, support, research, and coding. And companies don't want to run that AI themselves: they want vendors to deliver the savings inside tools they already trust. “Software as a service” becomes “software as a labor.”
2. Translation is the scarce layer. Calling an AI model is now trivial. Anyone can do it. Turning that call into a reliable improvement in a real business process is hard, and hard things command prices. “We'll just use ChatGPT” helps an employee. It doesn't change how an organization operates.
3. AI strengthens whoever owns the records. A common mistake is treating the interface as the product. The real asset sits underneath: the customer records, tickets, contracts, payments, and permissions AI needs in order to act at all.
Companies that own that layer become the thing agents have to plug into.
4. Pricing will move from seats to work, and that's survivable. Yes, seat pricing is exposed. But pricing pressure and demand destruction are different things. The vendors that adapt will charge for what the system does (tasks completed, transactions processed, outcomes delivered) instead of who logs in. The honest caveat: nobody has proven the new pricing captures as many dollars as the old one. That's the live experiment, and we list exactly how we're tracking it below.
5. The messier the industry, the safer the software. A generic AI can summarize a restaurant's sales report. It cannot easily replace the system that takes orders, routes them to the kitchen, processes payments, schedules staff, and reconciles the books. The more an industry's workflow touches the physical world, the harder it is for a chatbot to swap in.
THE PRICE THE BEARS HANDED US
Six months ago, the awkward question for this thesis was “Sure, but isn't all of this already priced in?” It was: software traded at some of the richest multiples in the market.
Not anymore. After the SaaSpocalypse, software trades at roughly 23x forward earnings, down from over 80x at the 2021 peak. Goldman Sachs CEO David Solomon called the selloff “too broad.” JPMorgan, Morgan Stanley, Morningstar, and Deutsche Bank have all flagged the same disconnect: prices say structural decline, yet fundamentals say record earnings.
So let’s find that gap.
THREE COMPANIES, THREE TESTS
I'm testing the thesis on three names, picked because they represent three different versions of the survival argument, and three different ways it could fail.
Source: Milk Road
ServiceNow: the workflow machine
ServiceNow runs the plumbing of big companies: IT requests, HR cases, customer service tickets, and security incidents. Exactly the kind of work where automation has to be trusted, approved, and auditable: the translation layer in its purest form.
The 2025 scorecard, per the company: $12.9B in subscription revenue, up 21%. Contracted future revenue of $28.2B, up 26%. A 98% renewal rate: customers almost never leave. And 603 customers now pay more than $5M a year, up from 420 two years ago.
ServiceNow's AI product, Now Assist, crossed $600M in contracted annual value in 2025. That's still less than 5% of the business, with the company recently increasing their target from $1.0B to $1.5B in 2026.
For now, the numbers prove the underlying workflow machine keeps compounding while the AI layer gets bolted on. The thesis gets tested by whether Now Assist hits that billion mark.
The bear risk: AI-native challengers chip at individual modules, and the company itself has warned that AI infrastructure spending will squeeze its gross margins, a reminder that AI raises costs a lot for software makers, too. But a challenger has to do more than build a better screen. It has to rip out the operational nervous system of a large enterprise: trust, integrations, permissions, audit trails, procurement. That's a much bigger lift than a UI.
My take: ServiceNow is the cleanest version of the thesis, and the 2026 Now Assist number is its first real exam.
Salesforce: the stress test
Salesforce is classic SaaS: customer records, per-seat pricing, enormous scale. If SaaS were truly dead, this should be the most exposed name on the board, which is exactly why it's the best test. And to be fair, the market is treating it that way: the stock fell roughly 50% during the selloff.
The latest fiscal year (FY26, ended Jan. 2026), per the company: $41.5B in revenue, up 10%. $15B in operating cash flow. Slower than the glory days, but nowhere near being classified as a “collapse”.
The number everyone quotes (and the one to read carefully) is the AI line: “Agentforce and Data 360” hit $2.9B in annual recurring revenue, up over 200%. Sounds explosive… but if you unpack it: $1.1B of that is Informatica, a company that Salesforce bought. The actual agent product, Agentforce, is about $800M, up a genuine 169%, but still roughly just 2% of company revenue.
Salesforce also started reporting a brand-new metric: “agentic work units,” as in tasks completed by its AI agents. It has logged 2.4B of them so far, growing 57% a quarter.
They might be trying to flaunt an AI-driven metric versus an antiquated customer-driven one, like “seats sold”. The old model was humans × subscription. The new one is workflows automated × value delivered. Nobody knows yet what a “work unit” is worth in dollars. But that’s the game they are playing here.
The bear risk is straightforward: CRM is mature, seat growth is fading, and the new AI revenue is still too small to carry the company. If agent revenue stalls while seat revenue erodes, Salesforce gets caught mid-jump.
My take: Salesforce is the live experiment in re-pricing software from seats to work. I'm watching one number: Agentforce ARR. It needs to keep doubling.
Toast: an industry's operating system
Toast matters because it carries the thesis out of the office and into the physical world. It runs restaurants: orders, kitchen screens, payments, payroll, loyalty, and inventory.
In practice, it works as an outsourced operating system for an entire industry.
The 2025 numbers, per the company: recurring revenue passed $2B, up 26%. Locations grew 22% to about 164,000, with a record 30,000 added in the year. It processed $195B in payments, up 23%. And here's the part that matters for the AI debate: Toast barely charges per seat at all. Roughly half its recurring revenue comes from payment processing: it earns when its restaurants earn. You can't “seat-compress” a payments stream.
Full disclosure: Toast is as much a payments company as a software company, and that is precisely why it holds up. A restaurant wants orders flowing to the kitchen, money reconciling, and staff showing up, not ten AI tools. Generic AI can help with pieces of that. It is not going to replace the machine that runs the whole show.
The bear risk here is economics, not AI. Restaurants are low-margin and recession-sensitive, payment volume per location dipped slightly last quarter, and hardware-plus-payments is a lower-margin business than pure software.
My take: Toast shows the endgame. In a vertical, software becomes the industry's operating layer and gets paid on volume, not seats. That model is largely AI-proof. Its risks are old-fashioned ones.
THE NUMBERS THAT DECIDE THIS
A thesis you can't prove wrong is just an opinion. So here is mine with tripwires attached: hit the bull numbers and I'm right, set off the tripwire and I’m wrong.
Source: Milk Road
That last tripwire is the big one. Klarna tried it and partially retreated. The day a major enterprise does it and sticks: that's the day the squeeze becomes the base case and I cut exposure to the incumbents fast.
WHAT WOULD CHANGE MY MIND
Beyond the tripwires, three slower-moving risks I take seriously:
- Agents get reliable faster than vendors adapt. Gartner says today's agents can't be trusted with complex goals. Models improve fast. If that gap closes in two years instead of five, seat revenue erodes before work-based revenue replaces it.
- AI melts the moat from below. Switching costs exist because migrations are painful. If AI makes rebuilding integrations and migrating data 10x cheaper, incumbents lose their best defense even if no single challenger wins.
- The new pricing captures fewer dollars. It's possible that “paying for work” simply nets out to less money than “paying for seats,” because AI deflates the price of the work itself. Survivable for the best, brutal for the rest.
WHERE WE LAND
AI raises the bar for what business software has to do.
The old model digitized work and charged for access. The new model automates work and charges for results.
That is a harder test, and plenty of companies will fail it: the thin dashboards, the seat-priced tools nobody can prove a return on, the products that were always just a screen between a human and a database. The market is right about those.
The companies that own the records, run the workflows, and can prove the productivity gain become the layer AI has to pass through. For the first time in years, you can buy them at a discount instead of a premium.
My position: I just added ServiceNow (NOW) to my portfolio at around $108.
And you will need to go PRO if you want to see my moves on the other two and whether I’ll actually hold NOW as my thesis plays out (or doesn’t). It’s just a buck for a seven-day trial that you can cancel anytime. You’ll also get access to my full portfolio.
I'll add more if Now Assist stays on track for its $1.5B 2026 target, and I'll cut if contracted future revenue growth falls below 15% (the same tripwires in the table above). A full ServiceNow report is coming soon.
The market just repriced all software as if it helps humans use tools. The money will be made in the software that replaces, coordinates, and amplifies human labor.
That software is entering its most important time, priced for its funeral.
AI-GENERATED PODCAST 🤖
We’ve turned this PRO report into an AI-generated podcast to make it even easier to digest. You'll find the audio player below. 👇️🎧️
Why ServiceNow, Salesforce, and Toast survive this new market
Disclaimer: This podcast was created using AI and is based on the research report above. While we've done our best to ensure accuracy, the audio may contain minor errors, technical glitches, or mispronunciations. Please note that this podcast provides an overview of the report and is not a comprehensive or definitive take on the topic.
BITE-SIZED COOKIES FOR THE ROAD 🍪
DeFi Saver turns Aave v4 into a trading bot for your loans: Stop losses, take profits, and one-click loan migration - we broke down the pros, cons, and fees so you don't have to.
Matt Hougan: Why did the market freak out when Strategy sold 32 Bitcoin out of 831,000? Because Saylor talks in infinities.
Melody Wright: Everyone keeps waiting for the Fed to cut rates and save housing, but they're watching the wrong thing.
The vacuum of Space (X): SpaceX’s $80B raise drains market liquidity, with $74B pulled from other risk assets.
Stablecoins + Soccer in Mexico City. Bitso Business is hosting LatAm's biggest stablecoin conference during the FIFA World Cup. Use code SC_15off_9nd4 to get 15% off tickets.**
*this is sponsored content.
Click here to skip the waitlist and join
MILKY MEMES 🤣
ROADIE REVIEW OF THE DAY 🥛
VITALIK PIC OF THE DAY
