Analysis

SaaSpocalypse: what Salesforce’s results and Agentforce do to the thesis

The thesis announced the end of enterprise software in the face of AI. Four days later, the company it named first posts its best results and partners with the firm that was supposed to bury it.

Office towers at dusk, their glass façades tinted red by the setting sun
Valuations fell; the software, meanwhile, is still running

Since the start of 2026 one phrase has been circulating: the “SaaSpocalypse”. It comes from Jefferies’ trading desk, where Jeffrey Favuzza used it to describe the abrupt correction in enterprise software valuations. The thesis fits in one question: why keep paying subscriptions when an AI can handle the customers, analyse the data and produce the documents?

The correction itself is no figment. On 30 January, Anthropic quietly pushed eleven plugins under the name Claude Cowork. Within forty-eight hours some $285 billion evaporated from software stocks worldwide. Six weeks later the loss approached one trillion; by April it reached two. Salesforce lost more than 40% of its market capitalisation year on year, Adobe close to 45%, and the vendors down by half are too many to list.

What the quarter answered

On 26 August 2026 Salesforce published its second-quarter results: $11.35 billion in revenue, up 11% year on year, above consensus. Full-year guidance was raised. The next day the stock closed at $252.10, up 22.6% in a single session — one of the largest one-day gains in its history.

The figure that contradicts the thesis most directly is not the revenue line. Robin Washington, president, chief operating officer and chief financial officer, reports that attrition — the share of customers who leave — remains near its lowest level on record, and that growth in net new annual order value is the strongest in four years.

In other words: the customers are not leaving. The exodus did not happen at the company it was meant to hit first.

What is MCP?

Part of this reversal comes down to a protocol. The Model Context Protocol, opened by Anthropic in November 2024, defines a standard way for an AI to reach a piece of software’s data and actions — without bespoke development for each tool. It is what lets an AI read a sales pipeline in a CRM or trigger an action in a team messaging app.

Adoption was fast: more than ten thousand active public servers and some 97 million monthly downloads of the development kits, against roughly two million at launch. On 9 December 2025 Anthropic handed governance to the Agentic AI Foundation, a directed fund of the Linux Foundation, co-founded with Block and OpenAI and supported by Google, Microsoft, AWS, Cloudflare and Bloomberg.

That governance detail is the most important fact in the whole affair. A protocol shared by competitors stops being one company’s advantage: it becomes infrastructure. And infrastructure does not replace the software it connects — it redistributes its value.

Three consequences for a management team

AI does not replace the company’s memory

An AI, however capable, needs somewhere to find the customers, the contracts, the invoices and the history of exchanges. That memory is organised, secured and governed — audit trail, access rights, compliance. Replacing the tool that structures it would mean rebuilding the same guarantees elsewhere. That is not nothing, and it does not happen by migrating to a prompt.

The interface, on the other hand, can disappear

This is the real movement. On 26 August, Salesforce and Anthropic announced Claudeforce: Claude becomes the default reasoning engine across the Salesforce ecosystem, and — this is the reversal that counts — a “Salesforce in Claude” plugin lets you update a pipeline or prepare a client meeting from within Claude, without ever opening Salesforce. It ships with thirty-seven prebuilt sales skills.

Agentforce is therefore no longer a layer of AI laid over a piece of software: it is one of the two directions of travel between the two worlds. Its annual recurring revenue now exceeds $1.5 billion, within an AI and data business approaching four billion.

Value moves from the interface to the infrastructure

If the way in becomes the AI rather than the screen, defensible value no longer lies in comfortable display. It lies in what remains indispensable when nobody is looking at the screen: structured data, business rules, permissions, integrations.

Conversely, software that merely displayed more attractively data available elsewhere loses its only reason to exist. A handsome interface over data you can query in plain language is not worth much any more.

It is not enterprise software that is under threat. It is the part of its value that rested on your having to open it.

What this changes for a company that buys software

The question to put to every live subscription is no longer “is this tool pleasant to use”, but: what would remain of its value if nobody opened its interface any more? What remains — the data, the permissions, the rules, the integrations — is worth paying for. What does not remain is up for renegotiation.

This is a matter of stocktaking and arbitration before it is a matter of artificial intelligence, and it is the ground Maeliom Consulting works on.

That question — what remains of a tool when nobody opens its interface any more — also arises inside organisations, in another form: what AI makes you gain shows immediately, what it makes you lose does not.

Common questions

What is the SaaSpocalypse?

The term, which emerged in 2026, describes the sharp fall in enterprise software valuations in the face of AI, based on a thesis: why pay for subscriptions when an AI can manage customers, analyse data and produce documents?

Will AI replace SaaS software?

Not the software, but the screen. The interface stops being the entry point; structured data, permissions, business rules and integrations remain essential and become the defensible value. Software that merely displayed data available elsewhere loses its reason to exist.

What is the MCP protocol?

The Model Context Protocol, opened by Anthropic in November 2024, defines a standard way for an AI to access a piece of software’s data and actions. Its governance was handed in December 2025 to the Agentic AI Foundation, within the Linux Foundation.

How should SaaS software be chosen in the age of AI?

By asking one question of every subscription: what would remain of its value if nobody opened its interface any more? What remains — data, permissions, rules, integrations — is worth paying for; what does not can be renegotiated.

Sources: Salesforce second-quarter fiscal 2027 results (26 August 2026); donation of the Model Context Protocol to the Agentic AI Foundation (Anthropic, 9 December 2025); analysis of the SaaSpocalypse (Forbes, February 2026).


Next article

What AI makes you gain shows immediately. What it makes you lose does not.

Read

A transformation to support?