Writing
A column, and the notes underneath it.
Two different things. The column argues; the notes show the working.
The Moneycontrol column
For Moneycontrol, on AI and the future of work — written for the people who have to decide, not the people who build. The full archive is on my Moneycontrol author page.
MoneycontrolWhat AI cannot do in an Indian office, and what to do about it
The playbook we have imported assumes that work sits in documents and calendars. In most Indian offices it sits in WhatsApp, in a phone call, and in somebody’s head.
Read the column02MoneycontrolIndian IT says AI made it faster. So why is it still billing by the hour?
Headcount and revenue have formally decoupled at India’s top IT firms. The industry is still selling time, appraising effort and grading colleges on placements — three metrics that AI has already broken.
Read the column03MoneycontrolYour first 30 days with AI: A guide for everyone who isn't a techie
You do not need to learn AI. You need to start working with it. There is a difference.
Read the column04MoneycontrolAI didn't replace engineers. It replaced the apprenticeship
The most important job AI is changing may be the very first one. The junior work AI absorbed was never really about output. It was how people learned to become senior.
Read the columnNotes
Shorter and more technical, published here. What broke, what it cost, and the thing I got wrong first — the working behind the arguments in the column.
Hours were never the product
Indian IT can be faster and still bill time, because time is what the contract knows how to buy. If the unit of sale does not change, the model is an overhead, not a redesign.
Read the piece02The junior ticket was the training
The first-year work AI absorbed was never really about output. It was how people learned to become senior. If you delete the ticket and keep the headcount target, you have not redesigned the craft. You have cut the ladder.
Read the piece03What actually breaks when you put agents in production
Agent frameworks demo beautifully and behave differently under load. The model is rarely the thing that stalls. The write is.
Read the piece04The work sits in WhatsApp
The imported playbook assumes work lives in documents and tickets. In most Indian offices it lives in a chat, a call, and somebody’s head. An agent that cannot see that will look busy and change nothing.
Read the piece05What the newsroom taught me about agents
Software could research, draft and assemble at real volume. None of it published itself. An editor still said yes. That is not a limitation of the old world. It is the design.
Read the piece06The store is the constraint
An agent that looks clever in headquarters and fails on a Tuesday in a store is the whole problem. Fashion already has owners. The platform has to sit on their work, not replace the org chart.
Read the piece07Stay in the room
The people who design the work should still be there while it is delivered. Handing a finished spec to a factory and leaving is how you get a system that cannot wait for a person.
Read the piece08A percentage on work you never handed over
Efficiency numbers that sit on work nobody actually gave to the system are a slide. Baseline the work as it is, then change it. A number without a before is decoration.
Read the piece09Why I publish a skill file
I spend my days writing files that say what a thing is for, when to use it, and when not to. It seemed only fair to publish my own, as a URL an agent can parse.
Read the piece10Put the approver in the first room
Risk in the last meeting is not governance. It is a veto with no design behind it. The person who signs has to help write what they are signing.
Read the piece11Baseline it before you change it
A number without a before is a slide. Measure the work as it is — waits, handoffs, who actually says yes — then change one write. Otherwise you will celebrate a tool and miss the queue.
Read the piece12The second use case costs as much as the first
Most organisations already have an assistant and an orchestrator. Almost none have a reusable constraint. That is why the next agent costs the same as the last one.
Read the piece13A write that cannot wait is not a hire
If you cannot withdraw the tools from every desk when a hire is paused, you do not have a hire. You have a plugin someone will find in six months.
Read the piece14Use the licences you already hold
Most stalled programmes do not need a new platform. They need a gate on the tools that are already paid for. I will not open with a sale when what you own will do.
Read the piece15There is no maintenance window
A platform that already has stores, readers or patients will not give you a Saturday. Design the cutover as if the busiest hour is the one you chose, because it often is.
Read the piece16Integrations are the work
A lending line inside a media company does not stall on the product surface. It stalls on partners, reconciliation and the people who have to live with a write they did not design.
Read the piece17The sample cannot support the grain
A forecast that sounds sure at postcode level is often just a thin sample with a confident voice. Publish the uncertainty with the number, or the first miss will teach you to.
Read the piece18Move the cache first
On a live news platform the application servers look like the migration. They are not. The cache and the warehouse are what make a no-window cutover survivable.
Read the piece