The SaaSpocalypse Is Half Right
On the second shape of software
One of my first jobs as a product manager was to mock up wireframes for developers to build from. I'd sketch the flow on paper or in whatever tool I could get my hands on, hand it over, and watch something recognisable as product come back a few days later, rough around the edges but functional.
The customer experience was nothing to celebrate, but the cycle from idea to running code was measured in days, and back then that felt like the best trade we could make.
When we started working with real designers, the product got better almost immediately. Screens had weight and flows made sense the first time a user touched them in a way my rough wireframes never could achieve. The cost, which nobody at the time would have called a cost, was that our development cycles started to stretch. What used to be an afternoon of sketching now travelled through rounds of mockups and internal review before engineering could write a line of code. Weeks turned into months. The products improved and the timelines slipped, and I learned to accept that as the price of craft.
For years that was the assumption every product team inherited. Good design took time, and the rest of the cycle politely waited for it while the designer held the first shape of the idea.
Over the past year though, I've watched that flow compress. Now we share problem statements with coding copilots and agents, and the screens arrive, sometimes badly and sometimes closer to what we wanted than the third round of mockups would have given us. It is the same pattern showing up across every creative function, which is why AI agents are reshaping creative workflows faster than most teams have updated their process docs.
On Friday, the market priced what I'd been watching. Figma's stock dropped as much as 7.28% after Anthropic shipped Claude Design, a tool built on Claude Opus 4.7 that generates prototypes, websites, slide decks and design systems from natural language prompts. Adobe and Wix fell in the same session. Commentators have been calling this the latest wave of what's been named the SaaSpocalypse, a repricing of enterprise software that has wiped more than $2 trillion from the sector since February.
And I think the SaaSpocalypse is half right.
The thesis behind the repricing is straightforward enough to follow. For two decades, enterprise software sold seats and every new employee at a customer meant another seat, another subscription, another line in the recurring revenue.
Salesforce, ServiceNow, Atlassian, Figma, and almost everyone in the IGV index were built on that assumption, and the market capitalised it generously.
But now, agents broke the arithmetic. If one AI system can do the work that previously required five analysts, a company renewing its software does not need five seats, or four, or three. It needs one, and possibly none, and it would quite like to replace the line item with API credits from the model provider directly. A Fortune 50 memo that leaked in January described plans to reduce Salesforce and ServiceNow licence spend by around 60% over the year, with the savings redirected to AI infrastructure. UBS cut ServiceNow to Neutral on April 10 and dropped the price target from $170 to $100, pointing to tightening software budgets at large enterprises as they redirect spend toward AI. The iShares software ETF is down around 30% this year through mid-April. More than $2 trillion in software market capitalisation has been wiped off since February.
I've spoken to consumers and business leaders this year who have already decided to stop buying software and start building their own agents to replace the tools they used to license, because the capability can now be assembled from the models directly and wired into the workflows they actually need. The procurement conversation has changed. The buyer has changed. In some categories, the entire premise of buying software has changed, and you can see it most clearly in the way internal tools are shifting from apps to AI.
So when Zscaler's CEO Jay Chaudhry went on CNBC last week and called the SaaSpocalypse overblown, arguing that AI is not reliable enough to replace critical business software, I don't think he was reading the room. His argument rests on AI not being good enough yet, which is an argument with a shelf life of about six months, and his customers can read a shelf life. The repricing is not investors panicking about a distant risk. It is investors pricing a mechanism that has already arrived in the budgets of the companies they hold.
But the story doesn't end there. And this is where I think the market is missing the second half.
What gets lost in the panic is that agents don't replace software. They need it. An agent that writes a contract still needs somewhere to store it, somewhere to check it against prior versions, somewhere to route it for signature, and somewhere to prove later that the right version was signed by the right person at the right time. An agent that designs a landing page still needs a component library, a version history, a staging environment, and a pipeline that pushes the code into production without breaking anything around it. What you end up with is a digital workforce sitting on top of your existing systems rather than a clean replacement of them.
The software does not go away. The user of the software changes.
For two decades, the user was a human with a login. Now the user is increasingly an agent with an API key, a task, and a budget of tokens. The interface the human used to navigate becomes a set of endpoints called through API integrations rather than a login screen. The seat licence becomes a metered bill. The feature that helped a human do the job faster becomes the function the agent calls to prove the job was done correctly. It's the same software, reshaped for a user who does not need a dashboard.
What the SaaSpocalypse is actually pricing, I think, is the end of software built for the first user. Figma built a product for human designers, and that product is now competing with an agent that can generate the output directly. Salesforce built a CRM for human sales reps, and that CRM is now competing with agents that can read emails, update records, and send follow-ups without anyone opening the interface. The per-seat economy priced those products on the assumption that the human user was essential.
Remove the human and the per-seat pricing collapses.
Very little of this exists at scale yet (software built for agents), which is why most of the commentary this week can only see the first half of the story. The software being priced down is the software built for humans in a world that no longer needs that many humans doing that work. The software that will be priced up, once the market catches up, is the software being built for the agents that are now doing it.
This is what I think matters most for anyone running a business right now, because the question has moved. It used to be "which SaaS tools will help my team work faster." It is now "which software will still be useful when my team is smaller and my agents are doing more of the work." Those are different questions with different answers, and answering the second one usually means accepting that your tech stack needs an agent layer rather than another tool.
And most procurement conversations haven't yet noticed they've moved. If you want a sense of the questions that should be replacing the old ones, the C-suite view of agents is a reasonable place to start.
There's an irony running through all of this, which is that the software likely to survive is the software considered unsexy for most of the last decade. The CRM that kept painful audit logs because some regulator demanded it. Or the legal platform that made you annotate which clause changed and why. Those are exactly the compliance workflows agents can actually operate inside, because the record of what happened is already part of the product.
The tools built for speed and simplicity, the ones that celebrated one-click actions and minimal friction, are less well-positioned than most of their founders would admit. An agent does not need to be delighted by a clean interface. It needs to know why it did what it did, and to prove later that it followed the rules. The software optimised for human convenience is being rewritten into software optimised for agent accountability, which in practice means governance decided before launch rather than after. The companies that already lived in that world because they had to are finding themselves with an unfamiliar advantage.
If I were renewing a SaaS contract this quarter, the question at the top of my list would be whether the tool has any way to cooperate with the agent I might hand it to next year, and whether the vendor is thinking about that question or still selling me seats. Running an AI audit of what you are actually paying for tends to answer it quickly.
I started as a product manager sketching wireframes because nobody else would. Then I learned to wait for designers because the product was better for it. Now I hand problem statements to agents and watch the screens arrive, sometimes well and sometimes badly, but always faster than the last model. Each version of that workflow needed different software underneath it to hold the work together.
The next version will too. We just haven't priced it yet.
All the Zest 🍋
Cien