From “Everything is a File” to “Files Are All You Need” How Unix Philosophy Informs the Design of Agentic AI Systems

Deepak Babu Piskala · 2026

Read From “Everything is a File” to “Files Are All You Need” How Unix Philosophy Informs the Design of Agentic AI Systems by Deepak Babu Piskala (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-11672.txt.

Introduction

Deepak Babu Piskala traces a Unix idea into modern agentic AI. The 2026 preprint runs from files to files-are-all-you-need. Unix's creed from the 1970s was everything is a file. That design philosophy collapsed diverse resources into uniform interfaces.

Piskala follows the idea through the DevOps movement. It then lands in present-day autonomous agent systems. The same principle solves like problems at new scales. Uniform interfaces let tools compose without custom glue. Agents that treat resources as files gain simple, sturdy tools. The essay links old operating system wisdom to AI design. It argues the file remains a powerful abstraction.

Catalog record

CREATORDeepak Babu Piskala
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE16 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-11672

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

From “Everything is a File” to “Files Are All You Need” How Unix Philosophy Informs the Design of Agentic AI Systems Deepak Babu Piskala Abstract This paper traces how a design philosophy from 1970s operating systems—Unix’s “everything is a file”—is finding new relevance in modern agentic AI. We examine the evolution from Unix through the DevOps movement to present-day autonomous agents, showing how the same principle of collapsing diverse resources into uniform interfaces continues to solve similar problems at different scales. For practitioners, the implication is clear: designing agents around file-like abstractions and code as the action language may offer a path to more maintainable, auditable systems. Figure 1: Evolution of the uniform abstraction principle: from Unix’s file system (1970s) through DevOps code artifacts (2010s) to agentic AI’s file-based context and memory (2020s). 1 Introduction In the early 1970s, Ken Thompson and Dennis Ritchie faced a practical problem at Bell Labs. The computing systems of the day required programmers to learn different interfaces for different resources—one set of commands for disk access, another for terminal control, yet another for inter-process communication. The proliferation of special-purpose interfaces made systems difficult to learn, compose, and maintain. Their solution was elegant in its simplicity: represent everything—devices, processes, network connections, data—as files. A hard drive and a terminal would both respond to the same operations: open , read , write , close . This wasn’t the only possible design, but it proved remarkably durable. The abstraction survived decades of hardware evolution and remains foundational in Linux, macOS, and countless embedded systems today [ 1 ] . Half a century later, builders of autonomous AI systems face an analogous challenge. An agent that needs to query databases, call APIs, manage cloud infrastructure, store memories, and interact with humans confronts a similar proliferation of interfaces. Each capability comes with its own protocol, its own authentication scheme, its own data format. Without some unifying principle, the complexity threatens to become unmanageable. What we’re observing now is a convergence toward solutions that echo Unix’s original insight. Coding agents like Claude Code and Cursor 1 1 1 Cursor: https://cursor.sh ; Claude Code: https://docs.anthropic.com/claude-code are increasingly treating filesystems as their primary substrate for context, memory, and action. Academic frameworks are proposing file-system abstractions for agent architectures [ 6 ] . The pattern suggests that the old lesson—uniform interfaces reduce complexity—applies as much to AI agents as it did to operating systems. This paper examines that convergence, tracing the thread from Unix through the DevOps era’s “everything as code” movement to today’s agentic AI designs. The goal is not to claim that files solve all problems, but to understand why this particular abstraction keeps proving useful and what that might mean for practitioners building agent systems today. 2 The Unix Foundation The Unix philosophy emerged from constraints. Early computing systems were expensive and complex, and the engineers at Bell Labs sought ways to make them more tractable. One key insight was that simplifying the interface to resources—even at the cost of some flexibility—could yield large gains in usability and composability [ 2 ] . The file abstraction accomplished this by providing a universal handle. Whether a programmer wanted to read data from a disk, receive input from a keyboard, or communicate with another process, they used the same set of system calls. The underlying complexity didn’t disappear, but it was encapsulated behind a consistent interface. The Unix Principle: Expose diverse resources through a uniform interface. Complexity is encapsulated, not eliminated. This design enabled composition. Because programs read from and wrote to file descriptors using the same conventions, their outputs could feed directly into other programs’ inputs. The Unix pipe—allowing commands like ps aux | grep python | less —became possible because all three programs spoke the same language of byte streams. Small, focused tools could be combined to accomplish complex tasks without any of them needing to know about the others. The abstraction also lowered the barrier to understanding the system. A developer who learned how files worked could apply that knowledge broadly. New devices and virtual filesystems could be added without changing the fundamental model. This extensibility helps explain why Unix derivatives remain prevalent: the core abstraction continues to accommodate new requirements. It’s worth noting what the file abstraction doesn’t do: it doesn’t eliminate complexity, and it doesn’t fit every use case perfectly. Device-specific control still sometimes requires escape hatches like ioctl() . But as a default model—a way of thinking about resources that covers most cases—the file metaphor has proven remarkably resilient. 3 The DevOps Era: Extending the Pattern Beginning in the 2010s, the software industry began applying a similar principle to operations. The movement that became known as DevOps, and later GitOps, treated infrastructure, configuration, policies, and workflows not as ad-hoc settings to be clicked through in consoles, but as code to be versioned, reviewed, and tested. The parallels to Unix are instructive. Just as Unix collapsed diverse device interfaces into file operations, the DevOps movement collapsed diverse operational concerns into code artifacts. The paradigm known as Infrastructure as Code (IaC) 2 2 2 IaC (sometimes abbreviated IaaC) treats infrastructure provisioning as software: servers, networks, and databases are defined in declarative configuration files rather than manually configured. —moved infrastructure definitions from manual console configuration to declarative files in Terraform or CloudFormation. Deployment workflows moved from runbooks to YAML configurations in GitHub Actions or GitLab CI. Security policies moved from human-interpreted guidelines to machine-evaluable rules in Open Policy Agent or Sentinel. Organizations that adopted these practices reported similar benefits to what Unix had provided: composability (pipelines could chain together), reproducibility (environments could be recreated from code), and auditability (version control tracked every change) [ 3 ] . HashiCorp articulated the principle explicitly: “Everything has to be code—whether it’s security, policy, automation, compliance, infrastructure—for all of the benefits we talked about.” — Seth Vargo, HashiCorp [ 4 ] The pattern extends well beyond infrastructure. Diagrams that once required manual drawing tools now emerge from text descriptions: Mermaid.js 3 3 3 Mermaid is a JavaScript-based tool that renders diagrams from markdown-like text definitions, enabling version-controlled documentation. and PlantUML generate flowcharts, sequence diagrams, and architecture charts from markdown-like syntax [ 7 ] . API contracts move from informal documentation to machine-readable OpenAPI and GraphQL schemas [ 8 ] . Database changes become versioned migration scripts. Data quality rules become executable specifications. Domain Traditional Code-Based Infrastructure Console clicks Terraform, Pulumi Diagrams Drawing tools Mermaid, PlantUML API specs Documents OpenAPI, GraphQL DB changes Manual SQL Flyway, Alembic Data quality Manual checks Great Expectations Presentations PowerPoint Slidev, reveal.js Table 1: The “as code” pattern across domains What unites these examples is the same move Unix made with files: replace heterogeneous interfaces with text-based representations that can be versioned, diffed, tested, and composed. 4 Agentic AI and the Return of the File Today’s agentic AI systems—autonomous agents that plan, execute, and adapt—face the interface proliferation problem in acute form. A capable agent might need to interact with REST APIs, SQL databases, vector stores, cloud consoles, file systems, web browsers, and human users, each with different protocols and conventions. Recent observations suggest that practitioners are converging on filesystems and code as the unifying abstraction. Jerry Liu documented

Read the complete text