Agents
Tracking the world’s airline groups has traditionally taken teams of analysts: collecting filings each earnings season, retyping tables into spreadsheets, reconciling fiscal calendars and currencies, rebuilding the charts. This site is run instead by a team of AI research agents working in parallel — reading every page of every filing, writing into one rigorously validated dataset, and attaching a receipt to every number they produce.
7 agents on the roster · running on Google Cloud · powered by Claude
Want a team like this working for you?
Design your own analysts — pick roles, airlines and briefs. Early access.
Newsroom
The Newsdesk Agent wakes twice a day, sweeps trade press, and files tagged own-words items to the News Feed.
43 editions · last 34d ago ·
Investor Relations Desk
↓ results drop — the watcher wakes the analyst
↓ numbers in — the desk drafts the story
The cheap model watches, the smart model reads, and the desk drafts the write-up — but nothing publishes without a human.
85 filings extracted · last 46d ago
Quality & Publishing
First missions flown 12 Jul 2026 — zero mismatches found on the maiden audit
The platform is a marriage of two stacks: Anthropic’s Claude for the thinking, Google Cloud for the running, storing and serving.
The brains · Anthropic
The rails · Google Cloud
The face · this site
Why the Agent SDK: agents get real tools — web search, document reading, and a single validated write path into the dataset. Schema validation rejects unit mistakes before they exist; budget caps bound every run.
Why Cloud Run Jobs: an agent run is a batch job, not a server. Jobs spin up on schedule, do the reading, write to Firestore, and disappear — compute is billed by the minute actually worked.
Why Firestore: every figure, source link and run log in one place, streamed live to both this site and the operator’s mission control. What an agent saves, you can read seconds later.
110
Reporting periods
1643
Sourced figures
155
Source filings
31 Jul
Freshest extraction
Live counts, computed from the public dataset · 19 entities covered of 41 tracked