I was speaking at a Microsoft event on AI last month when a senior professional from the audience asked a familiar question… but with a slightly different twist:
Where do you think AI will matter most inside a company?
My first answer was the obvious one… productivity gains, automation, better decision support, faster drafting, quicker analysis. All true… But I have been thinking about that question differently since then. While many of us are busy launching AI projects, are we framing them correctly?
The bigger shift, I believe, is this: AI is moving from being a better assistant to becoming a self improving loop. You define a goal, define a metric, set guardrails, and the system keeps testing, learning, improving, and repeating. That is why Andrej Karpathy recent “auto research” project ( https://lnkd.in/dSTyagfp ) felt like a signal of where things are heading. The technical details matter less than the pattern: an agent tries something, checks the score, keeps what works, discards what does not, and loops again.
And that is also where the risk begins.
The danger for companies is not too little AI. It is optimizing the wrong metric at scale.
A company can optimize for clicks and lose trust. For speed and lose judgment. For efficiency and lose resilience. Faster service with weaker judgment. Higher engagement with lower understanding. A weak metric once gave you a bad dashboard. Now it can drive a self improving system in the wrong direction.
Successful AI implementation in companies may depend less on having more AI, and more on having better judgment about what AI should optimise.
Many of you may already be thinking along these lines. Are you looking at your AI projects merely as better assistants… or as self improving loops designed with the right metric and guardrails?
Ilyas Khan and 48 others
Originally posted on LinkedIn on March 11, 2026.
Srinivas Jain 