How Frontier works.
Sources in, scoring by Jev, a written explainer out — every day.
What is Frontier?
A daily publication by Agentik OS that reads the frontier of AI engineering — arXiv, lab blogs, Hacker News, GitHub and what builders post on X — and explains what actually changed.
How are topics chosen?
Jev (TypeSafe) scores every item on five questions: is it really the frontier, how new it is, how much there is to explain, whether it will still matter in six months, and whether a builder can act on it.
Who writes the articles?
A writing model (gpt-5.1) drafts each article strictly from the sources we fetched, then Jev reviews it for grounding, depth, clarity and usefulness. Anything under the bar is not published.
How often?
Signals are refreshed every 6 hours, articles are published every day.
Can I trust it?
Every article lists its sources and separates what the sources establish from interpretation. Jev's grounding and depth scores are shown at the bottom of each article.
Sources we read
arXiv (cs.AI, cs.CL, cs.LG, cs.MA), OpenAI, Anthropic, Google DeepMind, Hugging Face, Meta AI, Mistral and LangChain blogs, Hacker News, new AI repositories on GitHub, and the posts our own AI, Vibe Coding and Dev radars collect on X.