The question to ask before building an AI product
Praneet Dutta built AI models at Google DeepMind and is now co-founder and CEO of Pomo, a company building AI growth agents for brands. Yuval Passov hosted him on the third episode of his podcast, and the central question raised there is deceptively simple.
Before building anything, ask: does the next model make us stronger. If the answer is yes, every new model release works in the product's favor, and every technological leap becomes a tailwind. If the answer is no, that is not bad news but a map: it points precisely at the area that requires focus.
Turning AI into your own critic
Dutta describes a second use that is practical and immediately available. He feeds the model transcripts of his own meetings and asks for an analysis of what he could have done better.
The detail that matters here is the instruction. He directs the model to be harsh and constructive rather than polite. Without that directive, responses tend toward affirmation and reassurance, and therefore carry little value. This is a practice anyone can begin today without special infrastructure.
Why a DeepMind engineer still decides on instinct
The surprise in the conversation was not technological. A person who built models at one of the world's leading research labs makes a considerable share of his decisions on instinct.
This deserves attention, particularly in organizations that expect AI to deliver complete certainty. At early stages sufficient data does not exist, and accumulated experience remains decisive. The tools change the pace of work, not the need for human judgment.
What this means for organizations building on AI
The three conclusions converge on a single position. First, build in places where model progress increases your value rather than eroding it. Second, high quality feedback is already available to anyone willing to ask for it correctly. Third, human judgment remains part of the equation even for those who understand the technology most deeply.