State of Agent Readiness (ARS) 2026 — Research Report & Open Source Suite

Glintbase Research & Benchmark Labs · ARS-1.0
Empirical benchmark of 100 AI engineering platforms evaluated across 1,000 Pathfinder scans and 450 live coding agent executions (Claude Fable 5, GPT 5.6 Sol, Gemini 3.6 Flash / Antigravity AI).

---

📊 Overview

This repository houses the complete open-source research manuscript, ReportLab PDF build engine, raw cohort datasets, high-resolution vector/PNG figures, and live coding agent harness execution reports for the State of Agent Readiness 2026 report.

Key Benchmark Discoveries

• Global ARS Mean: 50.7 / 100 (Grade C — Moderate Friction).
• Machine Entrypoint Deficit: 52% of platforms deploy /llms.txt, but only 34% provide reachable OpenAPI specs and 12% offer live MCP server manifests.
• The Agent Resilience Paradox: Modern LLMs succeed despite defective documentation by paying a massive token tax — costing enterprise teams $433,200/year per 10-developer pod on premier models (Claude Fable 5).
• Remediation Impact: Following the 9-Action Remediation Playbook boosts platform ARS scores by +22.4 to +35.1 points.

---

📁 Repository Structure

ars-report/
├── data/
│   ├── reports/
│   │   ├── STATE_OF_AGENT_READINESS_2026.md    # Full 8,000-word research manuscript
│   │   └── STATE_OF_AGENT_READINESS_2026.pdf   # Published 24-page PDF report (1.1 MB)
│   └── exports/
│       ├── full-scores.csv                     # Raw 100-platform cohort benchmark scores
│       └── research-dataset.json               # Complete 100-platform empirical dataset JSON
├── benchmarks/
│   ├── ANTIGRAVITY-REPORT.md                   # Google Antigravity Assistant (Gemini 3.6 Flash) benchmark
│   ├── CURSOR-REPORT.md                        # Cursor Agent benchmark report
│   └── OPENCODE-REPORT.md                      # OpenCode Interpreter benchmark report
├── scripts/
│   └── build_pdf_report.py                     # ReportLab PDF build pipeline & canvas renderer
├── public/
│   └── frames/                                 # Figures 1 through 10 (PNG & SVG)
└── README.md

---

🛠️ Compiling the Research PDF Report

The research report is programmatically built using Python and ReportLab with custom page templates, cover logo integration, header/footer canvases, page numbering, and embedded charts.

Prerequisites

• Python 3.9+
• ReportLab & Pillow

pip install reportlab pillow

Build Command

python scripts/build_pdf_report.py

The script compiles data/reports/STATE_OF_AGENT_READINESS_2026.pdf (24 pages, ~1.1 MB).

---

📄 License & Attribution

All research datasets, benchmark findings, and report manuscripts are published open-source under the MIT License.

When citing this research report in academic publications or engineering documentation, please use:

@techreport{glintbase2026ars,
  title={State of Agent Readiness 2026: Global Benchmark of 100 AI Engineering Surfaces},
  author={Glintbase Research & Benchmark Labs},
  year={2026},
  month={August},
  institution={Glintbase},
  url={https://glintbase.dev/research}
}