The $830 Billion Panic: What "Software Is Dead" means for the Business of Software?
Back in February, investors wiped about $830 billion off software companies, and the trigger was a handful of AI plugins for lawyers.
Here's the short version. Anthropic released legal-focused plugins for Claude Cowork. Investors pictured AI agents doing the work that whole SaaS products charge for, and according to Reuters, software and services stocks lost roughly $830 billion in market value after January 28. Thomson Reuters fell about 16% in a single day. RELX, the company behind LexisNexis, fell about 14%. People started calling it the SaaSpocalypse.
I work on an AI product for architects, engineers, and builders, so I watched that week with a very personal kind of curiosity. If software is dead, what exactly are we all building?
This is part one of a five-part series. Here I look at the business side. Part two is about users, part three is about us, the engineers. Part four is about developer relations, and part five is about accountability. Let's go.
What the panic got right
Some of the fear made sense, and it's worth naming those parts.
Per-seat pricing looks fragile. For twenty years, most SaaS companies charged per user per month. That model works while a person does the work inside the product. When an agent does the work of five people through an API, the company pays for one seat and gets five people's worth of output. Vendors have noticed. Salesforce and Microsoft already meter agent work with credits and agent licenses, and Salesforce has started moving big customers to flat agentic agreements.
Thin products feel exposed. A lot of horizontal SaaS is a nice interface on top of a database and a few workflows. If an agent can fill in the form, move the card and send the follow-up, the interface matters less, and for many products the interface was the whole product.
Buyers look for a different moat now. A beautiful UI used to be a reason to switch tools. Today the question in a sales call sounds more like "Can our agents work with this?"
What the panic got wrong
Analysts pushed back fast. JP Morgan's Mark Murphy called it "an illogical leap" to go from productivity tools to every company building its own replacements. Wedbush said "the Armageddon scenario is far from reality." Gartner described the plugins as a risk for task-level knowledge work and expects the systems that run critical business operations to stay in place.
And companies keep paying. Global SaaS spending is still on track toward roughly $315 billion.
The reason is boring, and it's important. Companies pay software vendors for much more than screens. They pay for:
- a system of record, the one place where the truth about customers, money and projects lives;
- accountability, a vendor who signs a contract, passes an audit and picks up the phone at 3am;
- risk transfer, so that when the payroll system makes a mistake, someone else carries part of the blame;
- integrations, years of connectors to the forty other tools a company runs.
An agent can click buttons all day. It still needs somewhere trustworthy to read from and write to. In a world full of agents, the system of record gets more valuable, because every agent depends on it.
A simple map: what dies and what lives
Here's how I sort it in my head.
| Layer | What happens | Why |
|---|---|---|
| Interface-only tools | Shrink fast | Agents can do the clicking |
| Per-seat pricing | Changes shape | Work moves from people to agents |
| Workflow glue | Gets squeezed | Agents can chain the steps themselves |
| Deep vertical products | Hold up well | Domain knowledge and data are hard to copy |
| Systems of record | Grow in value | Every agent needs a source of truth |
The pattern is simple. The closer a product sits to the screen, the more pressure its makers feel. The closer it sits to data, rules and responsibility, the safer it is.
What this means for those of us building products
If software is changing shape, we need to ask new questions.
Who is our user now? For many products, the honest answer is "a person and their agent." That makes a clean API, clear error messages and good docs as important as the UI. Agents read docs too, and they read them very literally.
What do we charge for? We can charge for seats, credits or outcomes, and each one has a catch. With credits, we give up some revenue predictability. With outcome pricing, the customer and the vendor have to agree on what success means, which is a hard conversation and a very healthy one.
Where is our moat? It sits in the data we hold, the trust we've earned and the domain knowledge we've built up. In my own work, the hardest part to copy is our understanding of how architects and engineers work: the documents they live in and the mistakes that cost them money. A competitor could rebuild our screens in a weekend. Learning that domain takes years.
So, is software dead?
Software as a set of screens people click through all day is shrinking. Software as the place where business truth, rules and responsibility live is growing.
The companies that panic about the first part will chase the hype. The companies that invest in the second part will become the thing every agent depends on.
In part two, I look at the same question from the user's side: what people want from software once an agent can do the clicking for them.
Which kind of software will be gone in five years, and which kind will you still happily pay for? I'd love to hear the takes in the comments.
Sources
- SmartScope: Structuring the "SaaS is dead" debate (Reuters figures, analyst quotes, Gartner view)
- The VC Corner: SaaS isn't dying, here's what is (SaaS spending)
- InformationWeek: Is SaaS dead, or becoming AI?
- Forbes: SaaSpocalypse is dead, the future of SaaS is SaaS
- Alumni Ventures: AI is eating software
Top comments (0)