AI News Digest – AI Agent for News Research and Automated AI News

Starting point

Working in the AI space means drowning in newsletter subscriptions, blog posts and LinkedIn content. Filtering out what matters took around five hours a week, without producing a structured overview.

Solution

The agent works in two stages. Both run every week, fully automated.

Stage 1: Digest – a sorted overview of the week’s newsletters

Every Friday, a Vercel cron job starts the pipeline via GitHub Actions. The agent:

1. retrieves the week’s newsletters from Outlook (MS Graph API),
2. clusters the content by topic using embeddings, so several newsletters on the same topic end up in one cluster,
3. removes duplicates, writes titles and rates every topic by relevance (1 to 5 stars),
4. researches the most important topics with Perplexity Sonar,
5. stores the digest and shows it in a dashboard with filters by week, category and relevance.

Stage 2: AI News – the top stories as researched articles

Later the same day, a second agent turns them into the AI News on this website:

1. Editing: An editor agent picks the stories that matter to companies from the highest-rated topics and sets reserve topics.
2. Research: Every story is researched again with Perplexity, limited to the week in question, so each article is based on primary sources such as vendor announcements and established media.
3. Writing: GPT-6 Sol writes every story in German and English, with a summary, key facts, business context and footnotes to the sources.
4. Quality check: Stories that the sources do not confirm, or that have fewer than two solid sources, are dropped and replaced by a reserve story.
5. SEO: The agent pulls search volumes from DataForSEO and uses them for each edition’s page title and description.
6. Publishing: The edition is committed to the website repository, Vercel deploys it, and IndexNow notifies search engines about the new pages.

The overview page lets you filter all stories by theme, company and week, with a theme heatmap and a network of the key players.

Research, summarization and preparation are typical tasks where AI agents deliver quickly. More use cases are covered in our practical guide to AI agents for companies (German).

Technology stack

• Vercel Cron and GitHub Actions (scheduling and execution)
• MS Graph API (newsletter retrieval from Outlook)
• Azure OpenAI: GPT-4o and GPT-5 models (analysis, titles, rating), GPT-6 Sol (editing and writing the AI News)
• text-embedding-3-small (embeddings / clustering)
• Perplexity Sonar (research with sources)
• DataForSEO (search volumes for SEO)
• Vercel Blob (digest storage)
• React, Vite and Tailwind (digest dashboard), Astro (AI News on the website)

Result & Impact

• Research time cut from around 5 hours to 10 minutes per week
• Extrapolated, around 250 hours per year (estimate based on 4.8 hours saved per week)
• Up to 10 researched AI news stories every week in German and English, with sources and no manual effort
• Running AI costs under €1 per week for research, writing and SEO data of the AI News
• Fully serverless: no permanently running servers
• Modularly extensible for client deployments, for example as an internal industry newsletter