On February 3, 2026, Salesforce, SAP, ServiceNow, Adobe, and Workday lost a combined $300 billion in market cap — in a single day. The big enterprise SaaS names closed down 7 to 11 percent that day.
The fear behind it was simple: what if AI agents just skip our software entirely? Three months later, SAP gave its answer. It locked down API access for any AI agent it hadn't explicitly approved. Meanwhile, Fivetran and dbt Labs were making the opposite bet — throwing their entire data layer wide open.
Why did SaaS stocks melt in a single day?
On Tuesday, February 3, 2026, the biggest enterprise software names dropped in unison. The reason was simple. As companies started handing real work to AI agents, the market got scared that the whole per-seat pricing model behind SaaS was about to break.
The numbers made it concrete. Forward P/E ratios for software companies compressed from 39x to 21x. Seat-based pricing's share of revenue fell from 21% to 15% over the year, while hybrid and consumption-based pricing rose to fill the gap, reaching 41%. Once agents started using software instead of people, charging "per head" stopped making sense — that's the read the market landed on.
People started calling it the "SaaSpocalypse". And it left every SaaS vendor with the same question: if AI agents can reach our data directly, what's our app even for anymore?
SAP chose to lock the door
SAP's answer was defense. In its updated API policy (v4/2026), Section 2.2.2 bans "interaction with (semi-)autonomous or generative AI systems that plan, select, or execute sequences of API calls" — unless it happens through SAP-approved channels. There are exactly four of those: Joule Agents, the SAP Integration Suite MCP Gateway, Business Data Cloud, and Agent2Agent (A2A) via the SAP Agent Gateway. Enforcement kicks in on June 9, 2026, via a security patch.
German SAP consultant Marian Zeis put it bluntly: "If you're restricted to documented APIs only, the vendor ends up controlling your future development". SAP CEO Christian Klein defended the move on an investor call, saying mass data requests or millions of API calls need to be throttled — framing it as protecting performance and domain know-how.
Worth noting
This isn't just an SAP story. On the a16z podcast, Fraser pointed to a fear shared across the vendor world: "my SaaS app has less value as an interface now that agents can reach the data directly." He called defensive API lockdowns exactly what they are — "declaring war on your customers".
ServiceNow moved the opposite way at the same time. It used MCP and the A2A protocol to build an orchestration layer that connects even third-party vendor agents. Facing the same crisis, one company locked the door and the other opened it.
The open side is winning
On the podcast, Fraser directly challenged the thing every SaaS vendor is afraid of — "data gravity." "Data gravity is completely fake. It's a myth that comes from bad data pipelines," he said. What actually matters, in his view, is context: "An AI agent without context is basically like using ChatGPT before it was connected to the internet".
On June 1, 2026, Fivetran and dbt Labs completed an all-stock merger. Combined, they serve over 100,000 data teams, with customers including OpenAI, Zendesk, Coupa, HubSpot, LVMH, and Pfizer. Fraser stayed on as CEO, with Tristan Handy becoming President. "The next generation of enterprise AI will be defined by the quality of the underlying data," Fraser said. Handy added: "Trust gets built at the infrastructure layer".
It wasn't just talk. They open-sourced dbt Core v2.0 (alpha) under Apache 2.0, and shipped Agents Schema, an open standard for AI agents, alongside it. Where SAP bet that locking data down means safety, Fivetran and dbt bet the opposite — that opening up is what earns trust.
| Closed — SAP's approach | Open — Fivetran, dbt, ServiceNow | |
|---|---|---|
| Agent access | Four approved channels only | Standardized via MCP and A2A |
| Customer reaction | Lock-in concerns, consultant pushback | Trust built via open source |
| Stated rationale | Protect performance and domain know-how | "An agent without context is useless" |
| Timeline | API block enforced June 9, 2026 | Merger closed + open-sourced June 1, 2026 |
The market is already moving this direction. As of an April 2026 survey, 78% of enterprise AI teams are already running MCP in production, and there are over 9,400 public MCP servers — with internal/private servers estimated at 28,000 to 37,000. Gartner projects that 40% of enterprise apps will include task-specific AI agents by the end of 2026. There isn't a lot of runway left for holding the door shut.
Is your company's data ready for AI agents?
This might sound like a story about giant vendors, but the underlying problem applies to every company. Here's a 5-step way to check whether your data is actually ready to serve as context for an AI agent.
- Map your data sources
List out where your CRM, accounting, help desk, and internal docs actually live. Scattered data means an agent can't build context. - Sync into a central warehouse
Use Fivetran-style ELT connectors to keep everything continuously synced into one place — Snowflake, BigQuery, Databricks, whatever fits. One-time exports don't cut it; you need ongoing sync. - Define and test it
Use a dbt-style tool to clearly define metrics like "revenue" or "active customer," and add tests. Without definitions, an agent will interpret your numbers however it wants. - Gate access through an MCP layer
Don't let agents wander freely — route access through a gateway with scoped permissions and audit logs. Blocking everything and opening everything are both wrong answers. - Check your vendor contracts' API policy
Before adopting a new SaaS tool, check whether it has an SAP-style "approved channels only" clause. Finding out after you've tried to connect an agent is too late.
Want to go deeper?
a16z Podcast — AI Agents and the Fight for Customer Data The full conversation between George Fraser and Martin Casado a16z.com
Fivetran + dbt Labs merger press release Full deal details and product roadmap fivetran.com
Breakdown of SAP's new API policy Section 2.2.2 in detail, plus industry backlash theregister.com
MCP enterprise adoption, mid-2026 Server counts and adoption stats andrew.ooo
SaaSpocalypse survival strategies Deep dive on the market drop and vendor responses samsungsds.com
