Transit: Print to digital, word for word

The problem

Why Transit exists

A print story rarely lives in one place: the headline and opening are on page 1, the rest jumps to page 7, and the photo sits with the photo desk. Most of a print edition never makes it online.

How it works

Seven stages, every word validated

01

Mirror & detect

A read-only mirror of the print system. Only stable files are processed; every story gets an ID and version.

1 / 7

What you get

Built for how desks actually work

Never changes the copy

Validation compares every word; any difference blocks delivery.

Failure tolerant

Recovers from a missing jump or an odd file and records every decision.

Editors stay in charge

Unsure? It asks. Sensitive categories are always held.

Learns safely

Learning improves skills and context, not the underlying model.

An AI harness compared with a fixed agentic workflow
	Fixed agentic workflow	Transit harness
First surprise	Breaks on a missing jump or odd file	Tries an approved alternative
When unsure	Guesses or fails silently	Asks an editor
Copy	May rewrite	Never changes a word
Record	Little or none	Every decision logged

Questions

Transit, answered

Last updated 27 October 2026

What is an AI harness?

A harness runs AI models inside your rules. It picks from approved skills, checks each result, tries an approved alternative when a step fails, and asks an editor when it is unsure.

Can it handle legacy Hindi fonts?

Yes. Converting legacy fonts such as Kruti Dev and Chanakya to Unicode is part of the parsing stage.

How is it rolled out?

In four stages: shadow (compared, not pushed), drafts with editor review, selective automation for low-risk sections, then expansion guided by measured edit rates.

Works well with