SaaS Apps Are Dying, but SaaS ERP Is Not
2025 was the year of LLMs. 2026 is the year of agents. And as more agentic platforms like Claude Code and Cowork shipped, $285 billion in SaaS market cap evaporated. The consensus formed overnight that AI is going to eat software, and the consensus is right, but only for certain types of software.
The SaaS products in trouble all look the same: they do one thing, they wrap it in a subscription, and the thing they do is now trivially reproducible by an AI agent. Scheduling tools, basic CRM, form builders, meeting transcription, tier-one support triage. These products existed because building and distributing software used to be hard, but neither of those things is true anymore. That part of the narrative is real, but then people make the leap to "AI will replace ERP," and that is where the analysis falls apart.
An ERP system encodes decades of regulatory requirements, tax rules, intercompany accounting logic, statutory consolidation across dozens of legal entities, and audit trails that have to survive federal examination. You are not going to vibe-code that. I have spent the last several years migrating data into Oracle Fusion Cloud, and the complexity is not in the UI or the APIs. It is in the business rules that took decades to encode and that auditors will not let you hand-wave away.
The distinction that actually matters going forward is not whether your vendor has AI, because everyone has AI now. The distinction is where the AI lives. A bolted-on AI connects to your platform through APIs, reads your data, maybe summarizes it, maybe drafts an email about it, but it is fundamentally a smart observer sitting outside the transaction and looking in. Think of a ChatGPT wrapper, a Copilot sidebar, or an MCP server that connects an LLM to your application's API. It can tell you what happened, but it cannot participate in what is happening.
An embedded AI operates inside the transaction flow. It processes the invoice rather than observing it. It forecasts the cash position using the same data model that the GL uses to post actuals. It inherits the same security model, audit trail, and data governance that the rest of the platform enforces. Oracle embedded more than fifty AI agents directly into Fusion Cloud applications this way, not as a premium add-on, but as part of the platform. AI Agent Studio lets customers build their own agents that run inside Fusion's security model. If your AP clerk cannot see a vendor's banking details, neither can the agent running in their workflow.
The reason Oracle can do this is that they are not just an ERP vendor. They are also a cloud infrastructure provider, a government cloud provider with FedRAMP High and DoD IL5/IL6 certifications, and an AI infrastructure provider running some of the largest GPU clusters in the world. When you own the full stack from compute to database to application, embedding AI into transaction workflows is a natural extension of the platform rather than an integration project. Most vendors have to partner across two or three of those layers and hope the seams hold. Oracle built them all.
This is obviously not an unbiased analysis. I work on Oracle Fusion implementations. But the point is not "pick Oracle." The point is that when you are evaluating platforms right now, the right questions are: Does the AI operate inside transaction workflows, or does it sit outside them? Is it governed by the same security model as the rest of the platform? Is it included in your existing licensing, or is it a per-user add-on that doubles your cost at scale? And when it makes a mistake, because it will, does the platform's audit trail capture what the agent did, or do you have to go spelunking through a separate system's logs?
The SaaS landscape is splitting into platforms with AI embedded in them, and platforms with AI bolted onto them. If you are evaluating enterprise software right now, make sure you are looking at the former.
If you want to hear more about migrating to Oracle Cloud, extending it, securing it, and using AI within it, come see SMX at Oracle Ascend.