Last updated: 11 August 2026
We use artificial intelligence in producing this publication, and we would rather tell you so plainly than let you work it out. This page sets out what it does, what it does not do, and who is answerable for the result.
We are not going to publish the internals. How an issue is assembled is our own work and it stays that way, in the same manner that the Methodology page explains what the tests are without publishing the code that runs them. What follows is about the division of labour, which is the part that affects you.
Claude, made by Anthropic, and Kimi, made by Moonshot AI. Both are used as engineering and research tools: they help build and maintain the software behind the publication, and they help gather and organise source material.
Sarvam AI, an Indian speech company, whose text to speech reads the audio edition aloud. The voice on that recording is a synthetic voice named Ritu. It says so itself, by name, in the first minute of every episode and again at the end.
If we add another tool, or drop one, this page changes.
It does not read the rotation graphs. Those are read by a person, by eye, one chart at a time. Which sectors are followed, which industry funds are shown, and which companies are carried into the tests is a human judgement, made by the author, and it is the single most consequential step in the whole process.
It does not decide what appears. Whether a company reaches an issue is settled by a fixed set of arithmetic tests applied to published figures. Given the same data those tests return the same answer every time, and they are stated in full in every issue. No model is consulted, and no model can overrule them.
It does not produce a single figure. Every number comes from a named data source or is calculated from that source’s figures by the arithmetic described on the Methodology page. Where a figure cannot be computed, a dash is printed. Nothing is estimated, inferred or filled in, by a model or by anyone else.
It does not write the company deep dives. Those are written by the author, from the filings, the earnings calls and the company’s own material. The judgements and the words are his.
This is worth setting out precisely, because “generated” has come to mean one thing and here it means another.
We write the sentences. They are written, edited and signed off by the author before they are ever used, and they are fixed in advance rather than produced on the day. What the software does each week is fill those sentences with that week’s figures. No language model writes, rewrites, or chooses any part of an issue at the moment it is published, and none decides what an issue says.
The audio edition works the same way. Its script is written and edited before recording, and the Sarvam voice reads it aloud without changing a word.
The Classroom at school.themicrocapminute.in is a free course, and AI does more there than it does in the newsletter. Its chapters carry audio narration read by Ritu, the same synthetic Sarvam voice, and every recording says so itself. The course also embeds explainer videos from the publisher’s YouTube channel: their scripts are written with AI help, the narration is Ritu, and the animation is programmatic. Every video is reviewed by the publisher before it is published.
Ayush Agrawal, the author and publisher. Every issue is his, and every error in one is his, regardless of which tool was involved in producing it. An AI tool is not a defence and we will never offer it as one.
Nothing here changes anything in the Disclaimer. This publication is impersonal commentary of general and regular circulation. It is not investment advice, it is not a recommendation to buy or sell anything, and it is not tailored to any person.
Questions about any of this are welcome at hello@themicrocapminute.in.