Frontier Operations

“If your AI agents haven't surprised you recently with their successes or failures, you are likely not operating at the frontier.”

Implementing the core pillars of the Frontier Operations framework (Boundary Sensing, Capability Forecasting, and Failure Model Maintenance) for IT Services and Information Analysis teams.

How it Works

1. Sensors (data/SOURCES.yml): A curated list of high-signal nodes on the internet (YouTube channels like AI Engineer, Anthropic's research RSS, subreddits, etc.).
2. Taxonomy (data/CAPABILITIES.yml): Your internal list of specific IT/Engineering skills you want to track (e.g., Complex React Hooks, Terraform Scaffolding, Scanned PDF Parsing).
3. Ingestion Engine (src/ingestion/): Python scrapers that pull transcripts and raw text from the Sensors.
4. Extraction Engine (src/evaluation/extractor.py): Uses OpenAI Structured Outputs to read the scraped text and extract concrete claims that map to your Capability Taxonomy.
5. Reporting (src/main.py): Generates an aggregated Markdown report (docs/CAPABILITY_REPORT.md). (A Next.js visual dashboard is currently in development in /dashboard).

Setup & Installation

Requirements

• Python 3.10+
• OpenAI API Key (For the Extraction Engine)

1. Python Environment Setup

# Clone the repository
git clone <repo-url>
cd <repo-dir>

# Create and activate a virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

2. Configuration

Create a .env file in the root directory and add your OpenAI API Key. If no key is provided, the Extractor will run in "Mock Mode".

echo "OPENAI_API_KEY=your_key_here" > .env

3. Run the Pipeline

To run the full ingestion, extraction, and reporting pipeline:

PYTHONPATH=. python src/main.py

This will generate/update the docs/CAPABILITY_REPORT.md file.

Project Structure

├── data/
│   ├── CAPABILITIES.yml      # The taxonomy of skills we track
│   ├── SOURCES.yml           # The list of sensors we scrape
│   └── extracted_claims.json # The database of aggregated claims
├── docs/
│   ├── ARCHITECTURE.md       # Detailed system design
│   └── CAPABILITY_REPORT.md  # Auto-generated aggregate report
├── src/
│   ├── ingestion/            # Scrapers (YouTube, RSS)
│   ├── evaluation/           # LLM extraction logic
│   ├── models/               # Pydantic data schemas
│   └── main.py              # The pipeline orchestrator
└── dashboard/                # (WIP) Next.js visual interface

The 5 Pillars of Frontier Operations

This project operationalizes the philosophy outlined by Nate B. Jones:

1. Boundary Sensing: Continuously defining exactly where the human-AI boundary sits for specific IT domains.
2. Seam Design: Structuring handoffs between humans and AI for clean recoverability.
3. Failure Model Maintenance: Documenting the specific texture and shape of how current models fail.
4. Capability Forecasting: Reading the "swells" of AI development to make 6-to-12-month bets.
5. Leverage Calibration: Treating human attention as the scarcest resource in an agent-rich environment.