From Artificiality This Week * Our Gathering: Our Artificiality Summit 2025 will be held on October 23-25 in Bend, Oregon. The
AI agents are evolving with orchestration layers enabling modular, autonomous systems beyond traditional SaaS, reshaping workflows, labor, and software design.
David Wolpert warns AI networks may evolve beyond human math, creating unpredictable, emergent intelligence that defies control and comprehension.
AI agents are evolving with orchestration layers enabling modular, autonomous systems beyond traditional SaaS, reshaping workflows, labor, and software design.
We like to map out the shifts in AI in a way that’s clear and practical. The landscape is complex, and no framework captures everything, but we’ve found it helpful to think about AI software in terms of layers. This is a simplified view—real-world boundaries are fuzzy—but we think it’s a useful lens for understanding where things are today and where they could be headed within a timeline you can plan around.
AI software is evolving rapidly as models transform into agents and systems, powered by a critical new layer of coordination software that has recently been termed the orchestration layer. It's a new form of middleware—positioned between foundation models and applications, serving as the coordination layer between them. In this sense, it functions as the operating system for agents.
Each layer adds something important: foundation models provide raw capability, orchestration layers make them usable across contexts, and application layers tailor them to specific needs. But, when it comes to agents, the real action—and disruption—is happening in the orchestration layer.
The orchestration layer is where modularity and adaptability are coming into focus. It’s new, and its capabilities are advancing quickly. Unlike foundation models, which are largely controlled by a few major players, orchestration tools give developers the freedom to build highly specific solutions.
Take Microsoft’s Copilot Studio. This tool allows users to connect models with external data, APIs, and workflows to create customized agents. These are systems that can adapt to different contexts and operate dynamically. Compare that to Salesforce’s Agentforce, which integrates deeply into Salesforce’s ecosystem but operates more like an enhanced feature set within a fixed workflow.
This distinction matters because orchestration platforms are enabling developers to create modular agents that can tackle vertical-specific tasks in new ways. These are more than smart workflows because they are designed to be the foundation for systems that can handle more autonomy.
The application layer, where companies like Salesforce and Canva sit, has been the backbone of enterprise software for years. These platforms made their name by empowering people—helping teams work more efficiently and scale their operations. But the current moment feels like a pivot.
Here’s the core tension: SaaS tools generate billions in revenue, but they’re tiny compared to the labor markets they support. Salesforce, for example, earns $35 billion annually, but it operates in a market where sales and marketing salaries total $1.1 trillion. Historically, SaaS companies targeted software budgets, but orchestration platforms—and the modular agents they enable—are looking at the much larger opportunity in labor budgets. Agents built in the orchestration layer aren’t designed to make people more efficient. They’re designed to take on the work directly.
What makes modular agents so compelling is their flexibility. Instead of relying on a single platform’s ecosystem, these agents can operate across systems, pulling data, calling APIs, and adapting their workflows as needed. Platforms like LangChain and Copilot Studio enable developers to orchestrate these systems, giving them a level of independence and adaptability that traditional SaaS platforms can’t match.
For legacy players, this creates pressure. Application-layer products are starting to look like a defensive play—deepening their integration with foundation models to keep users within their ecosystems. These tools are powerful, but they’re constrained by their ecosystems in ways modular agents are not.
This isn’t just a theoretical shift—it’s happening now, and it has clear implications for how companies plan for the future.
We’re already seeing signs of this transition, but the next few years will determine whether orchestration platforms can balance flexibility with controllability.
The orchestration layer is creating a new kind of flexibility and adaptability in software, and it’s forcing us to rethink what AI systems can do. At the same time, application-layer SaaS platforms are doubling down on their ecosystems, trying to hold their ground in a shifting market. These dynamics aren’t just about technology—they’re about how organizations allocate labor resources and how work gets done.
The systems being built and funded today in Silicon Valley will shape how agents and workflows evolve. We’re watching closely because this shift could significantly influence software design—and hence the technical architecture of the future AIOS—in the coming decade.
The Artificiality Weekend Briefing: About AI, Not Written by AI
