About
I get a business talking
to its own AI.
My name is Ahaan Singhania. I am twenty-four, I live in Mumbai, and I run a studio of eighteen people. Everything I sell as an AI system is already running that studio.
Most founders I meet already pay for AI. It writes their emails and it cannot tell them who is on leave tomorrow, because nobody ever gave it anything to look at. That gap is my whole job.
What I actually think
You don't have an AI problem.
You have a translation problem.
The tools are already good. The model on a founder's phone can read a contract, plan a week and write a report. What it cannot do is see the attendance register, the site photos, the WhatsApp group where the real decisions happened, or the sheet somebody keeps privately because the official one is wrong.
So the answer is almost never a new app. It is giving the AI something to look at, in a language the whole business already speaks, and then letting it do the dull half of the work. When that is done properly, nobody has to be trained. People carry on sending the same WhatsApp messages they always sent.
How I decide what to build
Six rules I will not bend.
-
01
AI goes only where it earns its keep.
Most of what a business does already works. Forcing AI into it adds steps and makes good people slower. I map everything, build where it pays for itself, and give you the list of things to leave exactly as they are.
-
02
You own it. You don't rent it.
The system runs on a computer in your office, on your accounts, with your data on your own drive. You get the machine, the code and the records. If we stopped working together tomorrow, none of it would stop.
-
03
The model decides. Code guarantees.
No scripted menus and no canned replies pretending to be an Agent. The model reads what was said and works out what is meant. Code sits underneath it making sure the booking is really made and the number is really right.
-
04
It asks before it acts.
Anything that changes a record, moves work or sends a message gets confirmed by a person first, and written down afterwards. An Agent nobody can audit is one nobody will trust twice.
-
05
If it needs training, it is wrong.
Eighteen people use our own system every day and not one of them sat through a session on it. It lives in the apps they already had open. A system that needs a manual gets abandoned in the second month.
-
06
I build it for myself first.
Attendance, leave, daily tasks, projects, files and our own sales pipeline all run on systems I built for my studio. I only sell what I already trust with my own payroll. It is also why I will tell you when something should not be built at all.
What that looks like
Four streams of work.
Every project on this site falls into one of these. Most of them you can open and run yourself, right in the page.
- Conversing AgentsYou message it like a person, and it does the work.7 built
- Agentic systemsIt watches something all day and brings you the decision.4 built
- Operations & dashboardsThe records a business runs on, kept without anyone typing twice.5 built
- Websites & immersiveSites and 3D people can walk through, with the AI inside them.5 built
The other half of my week
I also run a studio.
Cove Craft Studios is my 3D rendering and AI media studio in Mumbai. Eighteen people making walkthroughs of buildings that are not built yet, product films, AR and immersive work for property, retail and events.
It is a separate business from the AI systems, and it is also where they came from. Running it is what taught me which parts of a working day are worth automating and which parts are the job itself. The studio is the test bed. The systems ship from there.
Where I would start
Send me one report your team writes at the end of the day.
That one document tells me what to build, who it is for, and how much time it is taking now.