The Revision

TheRevision is an autonomous tech-news publication built around a single promise: open it once and you're caught up. It reads dozens of feeds across the AI and tech beat, recognises when a dozen outlets are covering the same development, and rewrites that story once — in a single, consistent voice — then publishes it. Around the clock, with no one in the loop.

The audience is the information-saturated professional who needs to stay current on AI and tech but has stopped opening news apps because the ratio of signal to noise stopped justifying the time.

The news industry optimises for attention, not comprehension. Headlines are written to make you click, articles are padded to fill ad slots, and the same story gets republished across twelve outlets with slightly different framings. Staying informed has become a full-time job most people quietly fail at — and nowhere is that worse than AI, where something genuinely changes every week and the coverage lands in a dozen near-identical versions.

The project started as a personal problem: keeping up with AI meant opening the same story ten times across ten tabs and de-duplicating it by hand. TheRevision is the fix — a systematic, model-driven publication that does the reading, collapses the echo, and leaves one clean account of what actually changed.

The product is built like a newsroom, not a prompt. Asking one model to "summarise today's news" produces fluent, forgettable output you can't debug. So the work is split across a chain of single-responsibility AI roles, each making one narrow decision: a triage step decides whether a story is even worth writing, a writer drafts it in the house voice, an editor votes to pass, rewrite, or discard — feeding its reasons back for another attempt — and a publisher is the final gate that checks a piece is complete, original, and ready before it ever goes live.

Stories cluster around events, not articles. When twenty outlets cover the same launch, TheRevision recognises them as one story by meaning rather than headline — catching the "buys" versus "acquires" rewrites that keyword matching misses — and publishes a single entry, not twenty rewordings. The publication's voice is enforced as a hard rule every piece must satisfy, not a tone the model is asked to approximate. And the system learns from being corrected: every editorial decision becomes signal that sharpens what it writes about next.

It's also agent-native. Every operator action — managing sources, tuning the voice, publishing — is exposed as a tool, so the newsroom can be run by a person at a dashboard or handed to an AI agent through the very same controls.

Live and running. The autonomous pipeline — discovery, semantic de-duplication, writing, editorial review, and publishing — operates unattended on an hourly cycle, with the AI and tech beat as its first surface.

The current focus is broadening the beat beyond tech and tightening the editorial guardrails that decide what the system won't write about, which matter as much as what it will. Reach out if you'd want early access, or if you have strong opinions about what a short version of the news should and shouldn't be.