Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

From Programs to Operational Systems

Beyond Procedures

For most of computing history, software has been procedural at its core. Even object-oriented, functional, and event-driven paradigms ultimately compile down to sequences of instructions executed by a processor. The program is a procedure. The runtime is an executor.

Agentic software demands something different. The software is no longer only a procedure — it becomes an operational environment for reasoning and action. The system does not just run code; it interprets goals, retrieves context, selects strategies, delegates tasks, validates results, and adapts.

This is a qualitative shift. The program becomes a system.

What Makes It Operational

An operational system is defined by its ability to:

  1. Accept intent — not just input, but goals with constraints and context
  2. Maintain state — working memory, long-term memory, and operational metadata
  3. Coordinate work — plan, delegate, parallelize, sequence, and synchronize
  4. Govern action — enforce policies, permissions, and boundaries
  5. Adapt over time — learn from outcomes, compress experience, evolve strategies

No single component delivers all of this. It requires a system — a composition of interacting parts with clear responsibilities, boundaries, and contracts.

The Progression

EraUnit of SoftwareRuntime Model
ProceduralFunctionExecute instructions
Object-OrientedObjectSend messages between objects
Service-OrientedServiceCall APIs across boundaries
AgenticOperational SystemInterpret intent, coordinate action

Each era introduced new abstractions to manage growing complexity. The agentic era requires its own.

Agentic Systems Need Runtime Structure

A language model with tool access is not an operational system. It is a powerful component without structure around it. Structure is what turns a component into a system:

  • A kernel that routes intent and coordinates reasoning
  • Processes that isolate work into bounded, manageable units
  • Memory services that tier, compress, and retrieve context efficiently
  • Operators that provide controlled access to external capabilities
  • Governance that enforces policy, permissions, and auditability

Without this structure, agentic software is fragile, opaque, and ungovernable. With it, agentic software becomes an operational system — composable, observable, and safe.