company · vision · building-in-the-open
Why we run our own company on the thing we sell
Dhanur AI's support, finance, operations, product and marketing are done by named AI employees, and every action and every rupee is on a public page. This is why we built it that way, what it has cost us to be honest about it, and what it is still not.
Every software company selling AI right now says it is AI-first. Almost none of them will show you what their AI actually did yesterday, or what it cost.
We decided to. Dhanur AI's support, finance, platform operations, product triage and marketing are done by named AI employees. A human founder supervises them. Every model call they make, with its cost in rupees, is written to a log, and a summary of their work is published on our company page, which refreshes from our live systems.
This post is about why we built it that way, because the reason is not marketing.
You cannot sell what you will not use
We sell AI employees to Indian businesses: agents that answer customers, qualify leads and handle routine work, with a person approving anything that matters.
An AI employee product whose own company will not trust it with its own support is not a product. It is a demo with a price. So the test we set ourselves was simple: if we would not let it answer our customers, we have no business asking a shop in Rajkot to let it answer theirs.
That turns out to be a demanding test. Our own agents have told a customer they could not see an image they had just read out. They have written formatting marks that reached a real chat window. They have drafted a social post announcing that we had no signups. Each of those was embarrassing, each was found because we were the customer, and each is fixed. None of them would have been found by a passing test suite — and we have 1,295 of those.
What "run by AI" has to mean to be worth saying
It is easy to claim. Here is the version we think is worth anything:
Named employees with one job each. Not "the system". Meera writes the daily report. Kabir watches the platform's health. Raghav handles support. Tara turns what customers ask into roadmap items. Ananya watches what everything costs. Ishaan drafts what we publish. Each has a short list of tools it may use and rules for when it must stop and ask.
A log you can check. Every model call and tool call, with tokens, latency and cost in rupees. "Run by AI" is a slogan until someone can audit it.
A person on anything that matters. A refund, a legal question, a custom price, a public post — drafted by an employee, released by a person. We do not think this is a limitation to be engineered away. It is the product.
Silence when there is nothing to say. Each employee's duty is written so that a quiet day produces no report, no inbox item and no cost. A dashboard full of manufactured activity would destroy the only thing that makes the log worth showing you.
What this costs us in honesty
Publishing what your company did every day is uncomfortable when your company has just started.
Our daily report for 16 September read, in full: "Zero activity across the entire platform. No signups, no new workspaces, no tasks executed." It was written by our own chief of staff, emailed to our founder, and summarised publicly. We did not remove it. A report you only publish on good days is not a report.
The same rule governs what we say about the product. Our feature status page marks things Live, Pilot or Next, and a thing is not Live until a customer can actually use it. WhatsApp is a good example: the code is built and tested, it has passed Meta's business verification, and we still mark it Next, because we do not yet have a production number. It is the first thing most Indian businesses ask us for, and we would rather say "not yet" than have you discover it after you sign up.
What you can actually use today
An agent that answers from your own material — your website imported and kept up to date, your price list, your FAQ — on your website, on a chat page with a QR code for businesses without a website, by email, or through our API. In English, Hindi or Hinglish, whichever your customer writes in. With approvals on whatever you choose, and the cost of every reply shown in rupees.
Prices are in rupees, pay as you go, from ₹50, with a GST tax invoice. Every new workspace starts with ₹25 of free credit, which is enough to try it on your own questions before deciding anything.
What this is not
It is not autonomous, and we are not trying to make it so. There is a person in the loop on anything that reaches the outside world, and a kill switch that stops all customer-facing AI work at once.
It is not a replacement for your team. The agents handle the questions that have answers, and hand over the ones that need a human — complaints, refunds, anything angry or unusual.
And it is not finished. We have no customers yet. Everything above is a thing we built and use ourselves, which is a real claim but a smaller one than a case study. When a business is using this in anger, we will write that post instead, and it will be more useful than this one.
Until then, the most honest thing we can offer is the evidence: what our AI employees did today, and what it cost.