# Autonomous Web Entities

> A white paper on autonomous web entities — self-operating digital agents that interact with web services autonomously.

[Paper page](https://vsl-landing-page-el7cn5p1w-ifly-leonards-projects.vercel.app/research/autonomous-web-entities)

## Metadata

- Authors: Leonard Selvaraja Fernando, Sooryah Prasath, Akshay Eti
- Year: 2026
- Organization: icrewsystems
- Type: White Paper
- Category: White Paper
- Status: Published

## Abstract

This white paper introduces the concept of Autonomous Web Entities (AWEs) — self-operating digital agents capable of interacting with web services, managing data pipelines, and making decisions within defined operational boundaries. As AI language models and browser automation technologies converge, a new class of persistent, goal-oriented web agents is emerging. We propose a formal architecture for AWEs comprising four layers: the Perception Layer (web interaction and data ingestion), the Reasoning Layer (LLM-driven decision-making with guardrails), the Execution Layer (action execution with verification), and the Memory Layer (persistent state and learning). Through prototype implementations at icrewsystems, we demonstrate AWEs handling customer support triage, automated data reconciliation, and scheduled reporting — reducing manual intervention by 87% for defined workflows. The paper also addresses safety, alignment, and the ethical considerations of deploying autonomous agents on the public web.

## Proposed Hypotheses

- H0: Autonomous Web Entities cannot reliably perform defined web workflows without exceeding acceptable error rates.
- H1: The four-layer AWE architecture enables autonomous completion of defined web workflows with >90% success rate.
- H2: AWEs reduce manual intervention by at least 75% for structured data reconciliation workflows.

## Table of Contents

1. Executive Summary
2. Defining Autonomous Web Entities
3. The Four-Layer Architecture
4. Prototype Implementations
5. Case Study: Customer Support Triage
6. Case Study: Automated Data Reconciliation
7. Safety & Alignment
8. Limitations & Future Work
9. Conclusion

## Data Collection Method

Prototype Development & Observation: Three AWE prototypes were developed and deployed in production environments at icrewsystems over a 6-month period. Success rates, error types, and human intervention requirements were logged and analyzed. Each prototype underwent 500+ workflow executions.

## Tools Used

- Google Gemini
- OpenAI ChatGPT
- Google Sheets

## Meta

Canonical LLMs file: https://vsl-landing-page-el7cn5p1w-ifly-leonards-projects.vercel.app/research/autonomous-web-entities/llms.txt