The Autonomic Journey: From Expert Systems to Enterprise General Intelligence

Abstract: For decades, computer scientists have sought to create autonomic computing systems that can configure themselves, fix their own bugs, and stay focused on specific goals. This journey began with the brittle logic of Expert Systems in the 1970s and evolved into the infrastructure-centric observe-analyze-act loops of the early 2000s. While successful at automating deterministic tasks, these iterations failed to scale across the non-deterministic “jagged frontier” of complex business logic. Today, the convergence of neural intent extraction and symbolic validation has realized this vision. We define this as Enterprise General Intelligence (EGI): a federated system of autonomous agents whose “world model” is grounded entirely by the real-time state of the Enterprise Knowledge Graph. In contrast to the unbounded, open-world nature of Artificial General Intelligence (AGI), EGI operates within distinct bounds and constraints.

We frame EGI as the Distributed Neuro-Symbolic Operating System of the Enterprise, tracing the lifecycle of an autonomous workflow through a 3-tier enterprise stack:

• Tier 1: System of Interaction (The Shell): The shift from graphical interfaces to intent-based interaction, where the OS dynamically interprets natural language and shifts the burden of navigating software from human to machine.
• Tier 2: System of Execution (The Kernel & Scheduler): Intent drops into a “Bring Your Own Agent” economy where specialized processes (e.g., HR, IT, Legal) use multi-agent reinforcement learning to negotiate cross-functional tasks. To prevent chaotic execution, an Orchestrator Kernel enforces governance and validation against guardrails. Before execution, the system acts as a CPU scheduler, simulating multi-step workflows to foresee errors in a sandbox. EGI abandons static automation graphs in favor of stochastic planning, semantic tool discovery, and dynamic replanning.
• Tier 3: System of Record (Memory & Drivers): Validated plans are committed to the enterprise state. To prevent hallucinations, the OS uses Knowledge-Grounded Retrieval, fusing RAG with Knowledge Graphs. Enterprise applications are analogous to “headless” device drivers, where the OS translates intent into standardized tool calling to update backend data invisibly.

Beyond architecture, we will address the empirical evaluation required for agentic systems, moving past linguistic benchmarks to metrics like Execution Success Rate (ESR) and latency/cost trade-offs in reflection loops. We conclude by exploring EGI’s ultimate trajectory: Ambient Intelligence. As software becomes invisible, autonomous plumbing, we present the next frontier of systems research: solving the verification bottleneck for multi-step trajectories, mitigating state drift, and designing HCI frameworks for humans to act as “Editors-in-Chief” governing a society of agents.

Bio: Sandeep Uttamchandani, Ph.D., is the VP of Enterprise AI at Palo Alto Networks, where he drives the company’s AI-first transformation by engineering its foundational platforms, Enterprise AI products, and vertical business solutions. His 25-year career spans from founding a garage startup to driving technical innovation at IBM, VMware, and Intuit. Sandeep is a recognized expert in the 0 → 1 → blitzscaling of AI and data systems, having authored the best-selling book The Self-Service Data Roadmap and filed over 45 patents in the field. He is a frequent speaker at major industry and technical venues, focusing on the critical balance between AI democratization and enterprise governance. He holds a Ph.D. in AI Expert Systems from the University of Illinois at Urbana-Champaign.