Overview
This is a conceptual, evidence-backed position: ideas are practical now, but the paper presents examples rather than large-scale evaluations.
Citations0
Evidence Strength0.50
Confidence0.70
Risk Signals11
Trust Signals
Findings with numeric evidence: 2/5
Findings with evidence refs: 5/5
Results with explicit delta: 0/0
Reproducibility
Status: No open assets linked
Open source: Unknown
At A Glance
Cost impact: 60%
Production readiness: 60%
Novelty: 50%
Why It Matters For Business
Treating resources as files and actions as code reduces integration work, makes agent behavior auditable, and lets teams reuse existing DevOps practices to manage agent artifacts.
Who Should Care
Summary TLDR
This short position paper argues that a decades-old Unix idea—treat diverse resources as files—reappears as a practical design principle for autonomous AI agents. The authors trace the idea from Unix to DevOps ("everything as code") and to modern agents that use file-like memory, code as the action language, and file-based context retrieval. For practitioners the paper recommends: expose resources via a file-like namespace, let agents generate code (with sandboxing) instead of bespoke tool integrations, and use DevOps tooling for versioning and auditing. The paper is conceptual and cites industry examples (Anthropic, Cursor) and frameworks (AIGNE).
Problem Statement
Modern agents must talk to many different systems—APIs, databases, vector stores, cloud consoles, UIs—each with different interfaces. That interface proliferation makes agents hard to build, maintain, and audit. The paper argues that collapsing these interfaces into a uniform, file-like abstraction plus code execution reduces complexity and improves composability and traceability.
Main Contribution
Historical analysis connecting Unix's 'everything is a file' to DevOps and current agent design.
Argument and evidence that file-like namespaces and code-as-action simplify agent architectures.
Key Findings
Practitioners are converging on filesystem and code abstractions for agent context and actions.
Code as an action language reduces the need for many custom tool integrations.
What To Try In 7 Days
Expose one external resource (API or DB) through a small file-backed adapter and let an agent read/write it.
Store conversation history and agent plans as versioned files in git to enable auditing.
Prototype one capability by having the agent generate code (SQL or Python) and run it inside a sandboxed runner.
Agent Features
Memory
Planning
Tool Use
Frameworks
Is Agentic
Yes
Architectures
Collaboration
Reproducibility
Risks & Boundaries
Limitations
Non-text formats need parsing before use with file reads.
Very large document collections require indexing beyond simple file search.
When Not To Use
Low-latency device control where spool-and-read is too slow.
Highly specialized binary protocols that don't map to textual files.
Failure Modes
Agent writes unsafe or insecure code that is then executed.
Stale or inconsistent files cause agents to act on outdated context.