Autonomous AI Creator is a full-stack AI-powered editorial system that can independently discover, evaluate, investigate, and create content from real-world AI and technology signals.
The system is designed around a self-driving editorial workflow while keeping humans in control of the final publication decision.
Content creators and editorial teams spend significant time searching for relevant topics, checking their importance, researching supporting evidence, deciding whether a topic is worth publishing, and preparing content.
Most automated content systems focus mainly on generation. They may produce content without sufficiently investigating whether the underlying topic is relevant, reliable, or actually worth publishing.
Autonomous AI Creator addresses this by combining discovery, evaluation, investigation, editorial decision-making, generation, human review, and learning into one continuous pipeline.
The system follows an autonomous editorial lifecycle:
DISCOVER → EVALUATE → INVESTIGATE → DECIDE → CREATE → HUMAN REVIEW → PUBLISH/REJECT → OBSERVE → LEARN
The agent discovers real AI and technology signals from external sources, evaluates them according to its configured persona, investigates promising topics using external evidence, makes an editorial decision, and generates a draft when the topic is considered publishable.
The generated content is then placed into a human review stage. Nothing is automatically published without explicit human approval.
- Real-time discovery of AI and technology topics
- Persona-based editorial evaluation
- Multi-stage investigation with supporting evidence
- Publishability decision with confidence and reasoning
- AI-generated social-media content drafts
- Human-gated content approval and publication
- Persistent autonomous run history
- Content lifecycle tracking
- Outcome recording for published content
- Learning and recalibration from real outcomes
- Transparent investigation evidence
- Failure isolation and terminal run states
- Production-oriented FastAPI + React architecture
Each pipeline run progresses through clearly defined stages:
- Discover – Collects signals from real external sources.
- Evaluate – Scores candidate topics for relevance, novelty, and significance.
- Investigate – Researches promising topics and collects supporting evidence.
- Decide – Determines whether a topic is publishable.
- Create – Generates content for publishable topics.
- Human Review – Places the generated content into a review queue.
- Publish / Reject – A human explicitly approves or rejects the content.
- Observe – Records real post-performance outcomes.
- Learn – Uses recorded outcomes to update editorial learning rules.
A completed run clearly reports its final decision, confidence, evidence count, generated content status, and next human action.
Autonomy does not mean uncontrolled publishing.
The system intentionally stops generated content at:
PENDING REVIEW
A human supervisor can then:
- Review the generated draft
- Edit the content
- Approve it
- Reject it
- Publish approved content
This creates a clear boundary between autonomous reasoning and human-controlled publication.
The system is designed around a transparency principle:
Every important value shown in the interface should trace back to real system state, database records, or external evidence.
The application avoids fabricated discoveries, evidence, metrics, or learning results.
External signals can come from sources such as:
- arXiv
- OpenAI
- Anthropic
- Google AI
- GitHub
- Hugging Face
- TechCrunch
- Hacker News
Investigations can use the Breeth knowledge/research service to obtain additional evidence.
The system does not stop after generating content.
Once content has been published, real performance outcomes can be recorded, such as:
- Impressions
- Likes
- Comments
- Clicks
- Shares
- Saves
- CTR
These observations are used to generate deterministic learning rules and maintain an audit trail of how the system's editorial behavior evolves.
- Python
- FastAPI
- PostgreSQL
- SQLAlchemy
- Alembic
- Uvicorn
- React
- TypeScript
- Vite
- CSS-based design system
- Persona-based editorial reasoning
- External research and evidence collection
- Content generation
- Outcome-based learning
The application is deployed as a production web service with the backend serving the frontend and API.
The interface is designed as an editorial mission-control workspace.
It provides:
- Command Center – Monitor autonomous runs and current decisions.
- Discoveries – Explore discovered signals and investigation evidence.
- Content Studio – Review, edit, approve, reject, and publish generated content.
- Intelligence – View learned editorial rules and outcome-based insights.
- Memory – Explore what the system has retained from previous activity.
The interface reflects live backend state rather than relying on fabricated demo data.
The pipeline is designed so that individual failures do not prevent the entire system from completing.
External sources have isolated failures and timeouts, individual topic errors are captured, and autonomous runs are designed to reach a terminal state rather than remaining indefinitely in an "investigating" state.
The goal of Autonomous AI Creator is not simply to build another AI content generator.
It is to demonstrate how an AI agent can operate as an autonomous editorial system:
finding signals → reasoning about them → researching them → making decisions → creating content → learning from outcomes
while maintaining transparency and human control over publication.
The complete source code, frontend, backend, database migrations, and AI development logs are included in this repository.
See PROMPTS.md for the AI-assisted development prompts and workflow used to build the project.