Charanya
Subramaniyan
Product & AI practice

How I work

Product management principles are easy to recite and hard to practise. These are the ones I actually use, shaped by shipping AI platforms into production.

01

Start from the decision, not the technology.

Every engagement begins with three questions: what decision is the client trying to improve, whose day changes if we succeed, and which constraints has nobody written down. I am suspicious of any feature I cannot connect to the answer. This also governs whether a problem needs AI at all: if rules or a well designed search box would do the job, I will say so, even when it is less exciting.

02

Treat trust as a product surface.

Probabilistic systems will sometimes be wrong, so I design for it rather than hoping a better prompt makes it go away. Provenance (every number traceable to its source), confidence ratings, visible citations, and human sign off where the stakes warrant it. The principle is simple: do not ask users to trust the AI, give them the means to check it.

03

Define the evaluation before the build.

A realistic test set from actual client material, agreement with real users on what good looks like, and explicit thinking about failure modes, because a wrong answer stated confidently is worse than no answer. The shipping bar depends on the stakes and the safety net: a draft a person reviews can ship at a lower accuracy than a number going into a board report.

04

Work as a partner to engineering.

Engineers join discovery early, decisions come back fast so nobody is blocked waiting on me, and scope is protected from drive by changes. I stay close enough to the system to ask real questions about how it behaves, not because I will fix it, but because a product manager who understands the mechanics makes better trade offs.

05

Build the AI native muscle.

I use AI coding tools to make and deploy application changes directly for small product driven work, and I prototype working software during discovery so stakeholders react to something clickable rather than a slide. I am clear eyed about the limits: I am not an engineer, the team reviews what I ship, and I do not touch anything architectural. The point is that AI collapses the distance between having a product idea and testing it, and product managers who do not build that muscle will be at a real disadvantage.

06

Write things down.

Decisions that are written down clearly do not get relitigated. Playbacks happen early, because being corrected in week two is cheap and being corrected in month three is not.