Databases were not designed for the agentic space

Arpit Bhayani

Arpit Bhayani

Mar 08, 2026 • 1 min read


Databases were not designed for the agentic space. It is not about supported data types and algorithms, but rather about the assumptions on which they were built.

All databases were designed assuming human-written applications, predictable queries, intentional writes, short-lived connections, and careful developers sitting in the loop.

I wrote an essay on how agentic AI is quietly breaking the implicit contract that databases were designed around.

In this article, I break down which parts of the database are failing, why it matters for reliability and safety, and what patterns we can adopt right now to make our databases better suited for AI-driven workloads.

If you are building systems where AI agents can read, write, or operate on databases, this essay will give you a useful mental model. Give it a read.

Arpit Bhayani

Principal Engineer II at Razorpay - building Agent Studio, Ex-staff engg at GCP Memorystore & Dataproc, Creator of DiceDB, ex-Amazon Fast Data, ex-Director of Engg. SRE and Data Engineering at Unacademy. I spark engineering curiosity through my no-fluff engineering videos on YouTube and my courses