The Signal in the Noise
This week, the market sent an unmistakable signal. Anthropic’s launch of Claude Cowork — an agentic AI workspace with cross-application connectors — triggered the largest single-day sell-off in SaaS history. Hundreds of billions in market capitalization evaporated. Software stocks that had been the backbone of enterprise IT for a decade hit all-time low multiples.
The headlines called it the “SaaSpocalypse.” But what actually happened wasn’t destruction — it was migration. Value didn’t disappear. It moved.
Understanding where it went — and why — is critical for any enterprise leader making AI investment decisions today.
The In-App AI Trap
For the past three years, every major SaaS vendor has rushed to embed AI into their products. Salesforce shipped Einstein. Microsoft launched Copilot across its suite. ServiceNow, Workday, SAP — everyone added an AI layer.
The pitch was compelling: AI that understands your data, embedded right where you work.
The reality has been underwhelming.
Here’s the problem: in-app AI is sandboxed. Salesforce’s AI can only see Salesforce data. Microsoft Copilot reasons over Microsoft’s ecosystem. Notion AI knows about your Notion workspace and nothing else. Each copilot operates in isolation, blind to the rest of the enterprise.
But enterprise work doesn’t happen inside a single application. A field service call involves the CRM, the knowledge base, the parts inventory system, the scheduling platform, and the customer’s service history spread across three different tools. A clinical decision requires data from the EMR, published literature, internal protocols, and regulatory guidance. A claims adjudication touches member records, provider networks, policy documents, and compliance rules across multiple systems.
No single-app copilot can reason across these boundaries. They can summarize what’s in their silo, draft responses based on limited context, and surface basic analytics. Useful, perhaps. Transformative? Not even close.
As Jason Calacanis put it on this week’s All-In Podcast, the in-app AI features from Notion, Slack, and similar platforms are “nice,” but agents that pull from calendar, email, documents, and messaging simultaneously are “unbelievable.” His team built an internal agent called “Ultron” that aggregates every employee interaction, document, and communication channel into a single intelligence layer. It doesn’t live inside any one tool — it orchestrates across all of them.
Where Value Is Actually Moving
Brad Gerstner shared Goldman Sachs data on the show illustrating a stark trend: traditional software profit pools are shrinking while the agentic orchestration layer — AI that coordinates and executes work across systems — is capturing exploding value.
This isn’t about better chatbots inside existing tools. It’s about a new architectural layer that sits above your systems of record and transforms them into something fundamentally more capable.
Think of it this way: your CRM, ERP, EMR, and knowledge management platforms are data stores. They remember things. The agentic layer reasons over that data, connects context across system boundaries, and takes action. It turns passive records into active intelligence.
David Sacks framed it precisely: the greatest threat to SaaS isn’t replacement — it’s value capture moving upward. AI copilots inside tools are limited to their sandboxes. Cross-system agents become the preferred workspace. SaaS risks becoming legacy infrastructure — still necessary, but no longer where the value accrues.
The New Enterprise Architecture
What’s emerging is a clear architectural pattern:
Infrastructure layer — Your existing systems of record. CRMs, ERPs, data warehouses, document repositories. They’re not going away. They still store critical data and run core transactions.
Semantic knowledge layer — An intelligence fabric that ingests, organizes, and connects knowledge across all those systems. Not flat retrieval (the RAG approach of chunking documents into vector databases), but structured semantic understanding — ontology-grounded, provenance-tracked, temporally aware.
Agentic orchestration layer — Autonomous agents that reason over the knowledge layer and execute workflows across system boundaries. They plan multi-step processes, coordinate actions, make decisions within governance bounds, and deliver outcomes — not just suggestions.
The value is concentrating in the middle and upper layers. And critically, these layers must be independent of any single vendor’s ecosystem. An intelligence layer controlled by your CRM vendor will always prioritize their data and their workflows. True enterprise intelligence requires vendor-agnostic orchestration.
What This Means for Enterprise Leaders
The investment implications are clear:
Stop doubling down on siloed AI features. Every SaaS vendor will ship AI capabilities. Most will be incremental. They’ll make individual tools slightly better but won’t transform how your enterprise operates.
Invest in the orchestration layer. The enterprises that gain competitive advantage will be those that deploy cross-system intelligence — a semantic fabric that understands their domain, connected to agents that can act across their entire technology landscape.
Demand governance from day one. In regulated industries, this orchestration layer must be governed, auditable, and deployable in your own environment. The agent that processes claims, makes clinical recommendations, or executes field service workflows needs to operate within explicit policy boundaries with full traceability.
Preserve your data advantage. As Gerstner noted, durable moats come from data depth, not application logic. Your institutional knowledge, domain expertise, and operational data are your competitive advantage — but only if they’re organized into a semantic fabric that AI can actually reason over.
The Opportunity
The SaaSpocalypse isn’t the end of enterprise software. It’s the beginning of a new era where intelligence — not just data storage — becomes the primary value driver.
The enterprises that recognize this shift and invest in governed, cross-system AI orchestration will operate at a fundamentally different level. Not just faster at existing work, but capable of work that was never possible before: real-time synthesis across millions of data points, autonomous execution of complex multi-system workflows, and continuous optimization that compounds over time.
The value didn’t disappear this week. It migrated to a new layer. The question is whether your enterprise is building there.