The Journalist
and the Machine

It begins, as the best stories often do, with a bad joke. Two old friends — Anand, a data scientist who moonlights as an "LLM Psychologist," and Rohit Saran, the Managing Editor of The Times of India — meet in a Singapore hotel on a February afternoon in 2026. Rohit has just come from a conference organized by the INMA and OpenAI for journalists. He barely says hello before saying:

And so two men who have talked about data journalism for fifteen years — who both know that the story is in the numbers, never in the noise — open their most consequential conversation ever by talking about body mass.

Anand had indeed lost 25 kilograms. Rohit had put some on. "Net difference is…" Anand begins. Rohit laughs. And then, inevitably, they get to work.

What happened over the next six weeks would quietly change something fundamental about how one of the world's oldest and largest English newspapers produces content. Not with a dramatic press release. Not with a CEO announcement. But through a series of video calls, WhatsApp voice notes, debugging sessions, and at least one memorable conversation about induction cooktops.

The World's Largest English Newspaper Has a Problem

The Times of India is enormous. Founded in 1838, it employs hundreds of journalists across India, publishes editions in over a dozen cities, and reaches millions of readers every single day. Its digital properties dwarf most Western news organizations. This is not a scrappy startup experimenting with AI. This is the establishment.

And yet.

Every morning, somewhere in the TOI newsroom, a reporter was sitting down to write "Hack of the Day" — a short, practical technology tip aimed at readers who might not know that they can check how many SIM cards are registered under their name, or lock their Aadhaar biometrics, or file their EPFO e-nomination before something goes wrong. Good, useful, entirely human journalism. Except that after four months of daily publication, the human well was beginning to run dry.

The problem was not quality. The problem was volume. And volume, it turns out, is exactly what AI is good at.

Singapore, February 12, 2026

Rohit Saran had come to Singapore for the INMA/OpenAI conference. He had spent a full day sitting across from the people who built ChatGPT, discussing how it might transform journalism. He was curious. He was skeptical. He was, by his own admission, only a light AI user.

Then he met Anand for dinner, and Anand said something he always says:

He opened a browser, dictated a prompt into ChatGPT — not because ChatGPT was the best model, but because it has the best transcription of the lot — then switched to Claude, and typed: "Find the MoSPI data. Download it. Analyze it. Create a nice data story with rich interactive visualizations about all the anomalies you can find. Run all the usual statistical tests — traditional, non-traditional — and give me a fantastic story."

The AI started writing code. It started searching. It found the data. It noticed that charts were broken and flagged them for repair. It caught anomalies no human had spotted in years of using the same government datasets — statistical impossibilities, impossible weather readings, a state spelled as "Maharashtra dollar" in a national crime report. Within minutes, it had produced a scrollytelling page that looked like something The New York Times would publish.

Rohit watched in silence.

The silence that follows is more eloquent than any statistic Anand could cite. In 20 minutes, an AI had done what would have taken a team of journalists and data scientists days. Not because it was smarter than humans, but because it was faster, tireless, and — crucially — it knew exactly what Anand had taught it to look for.

How Fast Is Fast?

In 2025, Anand had set himself a target: write 20 data stories by the end of the year. It seemed ambitious. By August, it became clear that it was not ambitious enough. It was laughably, embarrassingly conservative.

Think about what that means. A data story — the kind that requires downloading government datasets, cleaning them, running statistical analysis, finding the anomalous data point that changes the entire framing, writing engaging prose, building interactive charts — had historically taken weeks of skilled work. Now it was taking a morning.

Rohit, who had spent his career fighting for resources, time, and talent to do exactly this kind of journalism, understood immediately what he was hearing.

The Problem with Experience

Here is one of the stranger findings from Anand's months of AI work: the people who adopt these tools fastest are the newest arrivals. The interns. The people who have not yet calcified their workflows.

This is not, Anand is careful to note, a story about intelligence. It is a story about mental models. The experienced journalist has a very clear picture of how a story gets made — and AI does not fit that picture. The intern has no such picture, so they just use it.

Rohit, to his enormous credit, had none of this resistance. He had something better: he had seen what the tool could do, and he wanted it in his newsroom immediately.

The meeting in Singapore was supposed to be a reconnection between old friends. It ended as a strategy session. Anand would spend the next six weeks building something inside the Times of India's editorial pipeline. Not as a contractor. Not as a consultant. As something stranger and newer: an AI collaborator with a GitHub repository and a very long list of prompts.

The Vision: Five Phantom Reporters

Before they got to "Hack of the Day," Anand had a bigger idea. He called it "Five AI Reporters."

The concept was seductive in its simplicity: enroll five AI agents as phantom journalists inside the TOI newsroom. Give them the same evening news list that human reporters receive. Let them pitch stories every morning. And then — this was the crucial part — treat their output exactly as you would treat output from any junior reporter. Let the desk decide what to publish.

Each AI agent would have a different specialty. One would be a data anomaly hunter, scanning public datasets for things that don't add up. Another would be a visual storyteller. Another would specialize in consumer rights — scraping court judgments for stories about insurance companies that consistently reject legitimate claims, or the inexplicable gap in compensation awards across Indian states for identical smartphone damage.

This was not a small idea. This was a reimagining of what a newsroom could be — not as a collection of human journalists, but as a hybrid of human judgment and machine speed. The humans would still decide what to publish. But they'd have an infinite supply of pitches to choose from.

Rohit's response was characteristic of someone who had spent decades thinking about the future of journalism.

Enter: Hack of the Day

But big ideas need small entry points. You don't walk into a 187-year-old newspaper and replace the entire editorial floor. You find a door that's already open and walk through it.

The door was a small feature called Hack of the Day.

Richa, the TOI reporter who had been anchoring the feature for four months, was running out of steam. The format was simple — a daily practical tip about something digital in an Indian reader's life: how to check if someone has registered a SIM card in your name, how to lock your Aadhaar biometrics, how to file an EPFO e-nomination before it's too late. Useful. Real. Valuable. But finite. The well of obvious government hacks was nearly dry.

It was Saikat Dasgupta, TOI's editor and resident technology skeptic (who also happened to have built an internal AI tool called Beans used daily by all 400 TOI reporters), who first named it explicitly on a call in early March 2026:

90 more hacks. Three more months of content. Not written by a human. Written by an AI that Anand would build over a single weekend.

The Technical Journey: How a Machine Learns to Tip

The architecture of Hack of the Day is deceptively simple. It has four stages.