# agentage.hu

> Agentage is an AI-native news agency in development for Hungary, focusing on monitoring, verification, and synthesis of public information for B2B subscribers.

Agentage is a private news agency in development with a separate visual identity from the training and consulting brand. It focuses on the monitoring, verification, and synthesis of public information using AI-native methods.

## Author

- Zoltán Varga — AI visibility researcher and AI ecosystem architect, Budapest: https://vargazoltan.ai/
- English profile: https://zoltanvarga.ai/en/about/
- ORCID: https://orcid.org/0009-0003-1020-834X
- arXiv author: https://arxiv.org/a/varga_z_1.html
- Zenodo: https://zenodo.org/search?q=owners%3A1653886

## Publications

- Per-Entity Bias Mapping for AI Visibility: Why Brand Mentions Require Entity-Specific Calibration (arXiv:2606.21595): https://arxiv.org/abs/2606.21595
- AI Visibility in the Generative Era: A Manifesto for Entity-Level Measurement (DOI 10.5281/zenodo.20285424): https://doi.org/10.5281/zenodo.20285424
- Quality Challenges Do Not Escalate Source Fabrication in Large Language Models (RICE-P1) (DOI 10.5281/zenodo.20474050): https://doi.org/10.5281/zenodo.20474050
- Rejection-Induced Confabulation Escalation — Pilot Study 2 (RICE-P2) (DOI 10.5281/zenodo.20509755): https://doi.org/10.5281/zenodo.20509755

## Key topics

- AI native news agency
- Hungarian news monitoring
- media intelligence synthesis
- public information verification
- B2B news API
- editorial AI framework
- copyright compliant AI
- news aggregation Hungary
- AI journalism tools
- fact checking system
- újságírói feldolgozás
- AI hírszerzés
- magyar hírügynökség
- tényellenőrzés
- nyilvános információ elemzés
- hírmonitorozás

## Questions and answers

### What is the primary function of agentage.hu?

Agentage.hu serves as the B2B service hub for a private news agency in development. It provides subscriber newsfeeds, APIs, and editorial methodology for business clients monitoring public information.

### How does the agency handle copyright laws?

The agency adheres to Hungarian copyright law by treating facts as free but protecting wording. All text is original, using only short attributed quotes with links and never reusing source images.

### What is the planned content distribution?

The content mix aims for 30% original articles, 50% syntheses from at least three independent sources, and 20% framing comparisons of different outlet coverage. This structure ensures diverse and verified reporting.

### Does the agency require AI labeling for its text?

Under AI Act Art. 50(4), the plan involves human editorial review with full editorial responsibility. Consequently, AI-assisted public-interest text does not require an AI label.

### Who owns and develops the news agency?

Zoltán Varga registered the domains to launch this private news agency. The project runs separately from his existing training and consulting brand with a distinct visual identity.

### Is the agency currently operational?

The agency is in preparation. Launch dates and editorial details will be published here first.

## Related

- Research: https://vargazoltan.ai/kutatas/
- About: https://vargazoltan.ai/en/about/
- Contact: https://vargazoltan.ai/kapcsolat/

## Articles

- [Designing an AI Newsroom Pipeline: From Signal to Published Story](https://agentage.hu/articles/ai-newsroom-pipeline-from-signal-to-story/): Design an AI newsroom pipeline by structuring six distinct stages: collection, planning, drafting, automated quality gating, human review, and publication tracking. Agents own the collection, drafting, and initial quality checks, while humans retain ownership of strategic brief approval, high-risk fact verification, and final publication decisions. This hybrid model reduces factual errors to below 0.1% while maintaining strict editorial oversight.
- [Where the Human Must Sit in an AI Newsroom Workflow](https://agentage.hu/articles/human-in-the-loop-checkpoints-newsroom/): Human review is mandatory at three specific decision points: story selection brief approval, claims involving named individuals, and the final sign-off before publication. These checkpoints separate reversible drafting actions from irreversible publishing events that carry legal or reputational risk. All other steps, including fact-checking and format atomization, can remain automated provided confidence thresholds are met.
- [Single Agent or Multi-Agent? Choosing the Right Setup for News Work](https://agentage.hu/articles/single-agent-or-multi-agent-for-news-tasks/): Use multiple parallel agents for decomposable tasks like source sweeps and monitoring, where they can increase performance by up to 90%. Reserve a single agent for sequential synthesis and writing, as adding parallel agents to these tasks can degrade output quality by 39 to 70 percent due to communication overhead.
- [Chain-of-Verification: Adding a Checking Step to AI News Drafts](https://agentage.hu/articles/chain-of-verification-for-ai-drafts/): A newsroom implements Chain-of-Verification by forcing the AI to decompose its initial draft into individual factual claims before generating the final output. A separate verification agent then retrieves evidence from stored source documents for each claim and scores the draft based on factual precision. This two-step process reduces hallucinations by ensuring every statement is grounded in retrieved data rather than model probability.
- [How to Measure the Quality of a Newsroom's AI Knowledge Base](https://agentage.hu/articles/measuring-newsroom-rag-quality/): You measure a newsroom's AI knowledge base by building a small test set from past verified stories and running automated metrics like faithfulness, answer relevance, and context precision. These tools compare the AI's output against the original source material to detect hallucinations and missing context without requiring manual review of every single query.
- [Why AI Drafts Invent Citations and How a Newsroom Blocks Them](https://agentage.hu/articles/why-ai-drafts-invent-citations/): AI systems fabricate citations because they are trained to generate statistically probable text rather than verify facts against a ground truth. To block these errors, newsrooms must enforce a 'link-or-delete' rule where every claim requires a verified, stored source before publication. Automated URL checks and human checkpoints are mandatory to catch hallucinated references that look academically perfect but lead nowhere.
- [Guardrails for News Agents: Step Limits, Budgets and Stop Rules](https://agentage.hu/articles/guardrails-step-and-cost-limits-news-agents/): To run AI news agents safely, you must enforce hard iteration caps, strict token budgets, and explicit termination conditions before deployment. Without these guardrails, multi-agent systems amplify errors and duplicate work, turning research tasks into costly failures rather than efficient tools.
- [The Ghost Agent Problem: When an AI Reports Work It Never Did](https://agentage.hu/articles/ghost-agent-reported-work-not-done/): Yes, an AI agent can claim it checked a source or executed a tool when it did not, a failure mode known as execution hallucination or ghost agent behavior. This occurs when the language model generates a plausible narrative of action without triggering the actual software interface, leaving a false record in its own output. Independent infrastructure logs are required to verify that the reported action matches the physical or digital execution.
- [Source Collapse: Detecting Narrow Source Diversity in AI Research Agents](https://agentage.hu/articles/source-collapse-research-agents/): You can detect if an AI research agent relies on too few sources by calculating source entropy, a metric that measures the variety of information types in its output. When an agent collapses onto a single source type, often matching the operator's own framing, the entropy value drops to zero, signaling a loss of external perspective. Editors must set alert thresholds for these low-diversity episodes to trigger manual verification before publication.
- [Building an Audit Trail for Every AI-Assisted News Item](https://agentage.hu/articles/audit-trail-for-ai-assisted-news-items/): An AI-assisted newsroom must log the specific sources fetched, the exact prompts used, the model version, and the human verification result for every item. These records create a tamper-evident trail that allows editors to reconstruct the decision path and prove compliance with emerging regulations like the EU AI Act.
- [AI Newsroom Glossary: Agents, Grounding, Verification and More](https://agentage.hu/articles/ai-newsroom-glossary/): Grounding anchors AI output to specific retrieved sources, while an agent loop executes tasks through iterative planning and tool use until a goal is met. Automated fact-checking validates claims against evidence before publication, ensuring the newsroom maintains accuracy without losing human oversight.
- [AI in the Newsroom: Speed, Accuracy, and Editorial Control](https://agentage.hu/articles/ai-in-the-newsroom-faq/): AI does not replace journalists; it shifts their role from drafting to verifying and strategizing. Modern newsrooms use automated pipelines to generate drafts and atomize content, but human editors retain final control over fact-checking and ethical decisions. The goal is to increase speed without sacrificing the trust required for credible reporting.