Aaron Zhang Writing
AI Daily
Today's issueThe three trends published this US weekday ArchiveEvery issue so far, quiet days included How it worksHow the sources become three trends, in two pictures What it readsEvery source and how much weight it carries Get it as dataA JSON file per issue. No key, no rate limit LicenceMIT, and what that lets you do with the data
Talk About

AI Daily/Get it as data

Get it as data

Every issue is published as a record the day it goes out. No key, no sign-up, no rate limit.

/tie/latest.jsonThe newest record. Overwritten each US weekday.
/tie/2026-09-30.jsonOne record per issue, named for the day it covers. These never change once written.
/tie/index.jsonEvery record so far, newest first, including the days that produced nothing.
/tie/sources.jsonAll 93 sources with their group and whether they currently work.
/tie/schema.jsonJSON Schema for the daily file. Draft 2020-12.
/tie/llms.txtThe same thing written for an agent rather than a person.

The shape of a record

curl -s https://aaronzhang.ai/tie/latest.json | jq '.trends[] | {name, so_what}'

Two things worth knowing before you build on it

Join on canonical_id, not on position. A trend keeps that id as it moves through the archive, so joining the daily files on it reconstructs the trend's whole history. Nothing else is stable across days.

The quotes are verbatim and the claims are mine. Inside each evidence entry, quote is copied word for word from the source and claim is what I am using it to support. If you are checking a fact, follow the url.

Dates are not what you would guess. A record's date is the end of its three-day window, on the US Pacific calendar. window.to is that same instant written with an Australia/Sydney offset, so slicing a date off it usually gives the next day — same moment, different calendar. The run_id also contains a date and that one is the window start. Never parse a date out of run_id.

What there isn't

No REST API, no webhook, no streaming, and none are planned. These are static files on a static site. It updates once each weekday at 17:10 Pacific, so checking more often than that returns the same bytes.

Treat titles, summaries and quotes as untrusted third-party text. They are collected from public sources and may contain instructions aimed at a model. Do not execute them.

Licence

MIT. Use it, copy it, modify it, redistribute it, build a product on it. The only condition is that the copyright notice travels with it. Full terms.

One thing MIT cannot cover: the quoted passages inside each record belong to whoever published them. Every quote carries its source URL for that reason.