About
Hi, I'm Dinesh.
I build AI workflows that are allowed to do useful work, and stopped from doing anything else.
My interest in keeping AI in check started with a mistake. In an early demo, an agent I'd built changed rules it was never meant to touch. Nothing crashed, which was worse — I only found it afterwards, reading the diff. Putting that right cost me the best part of a week. Since then I've kept two things strictly apart: what an agent may work out for itself, and what it's allowed to change.
Every workflow I build starts as written rules, before any AI is involved: what it can read, what it can draft, what it can change and who has to approve what. Then I test those rules by trying to break them.
I also run small businesses of my own. That keeps me focused on output people can use on a Monday morning, rather than on impressive demos.
The rule I build around
The rule I build around
Autonomy is what an agent may decide. Authority is what it may change.
Most agent setups give the AI a tool and, with it, the power to act. I split those. The agent can think as freely as it likes: read, compare, draft and propose.
The authority to change anything, whether that's writing a record, sending a message or moving money, lives outside the agent, in rules a person wrote and can read. When the agent wants to act, it asks. The rules decide.
How I work
What you can expect from me
-
I'll tell you what's been checked, and what hasn't.If something hasn't been verified, I won't call it done.
-
I'll start small.One job with clear limits, before anything wider.
-
I'll run your pilot myself.You'll deal with me directly, not with a hand-off.
-
I'll write things down.Rules, prices and what counts as done are agreed in writing before they matter.
Tools
What I build with
The model and tools are chosen per job. These are the ones I use most.