---
title: "Sujal Maiti (@believer652) | Hashnode"
description: "Sujal Maiti (@believer652) writes on Hashnode. Read their latest articles and follow their work."
url: "https://hashnode.com/@believer652"
---

@believer652

# Sujal Maiti

@believer652 · Joined February 2022

### [I Think We're About to Have a Distributed Systems Problem Disguised as an AI Problem](https://sujal.hashnode.dev/i-think-we-re-about-to-have-a-distributed-systems-problem-disguised-as-an-ai-problem?utm_source=hashnode&utm_medium=feed)

We've taught agents to work together. I'm not sure we've figured out how to stop them from getting in each other's way. I've been spending quite some time experimenting with agentic workflows lately.

8 min read · [sujal.hashnode.dev](https://sujal.hashnode.dev/i-think-we-re-about-to-have-a-distributed-systems-problem-disguised-as-an-ai-problem?utm_source=hashnode&utm_medium=feed)

[6d ago](https://hashnode.com/posts/i-think-we-re-about-to-have-a-distributed-systems-problem-disguised-as-an-ai-problem/6af3f752fc4be55ac012e1ac)

### [Your AI Agent Said It Completed the Task. Did It Actually?](https://sujal.hashnode.dev/your-ai-agent-said-it-completed-the-task-did-it-actually?utm_source=hashnode&utm_medium=feed)

Why evaluating an AI agent's response is not the same as evaluating whether it successfully completed the task. Imagine asking an AI agent to schedule a meeting with your engineering team. It searches

15 min read · [sujal.hashnode.dev](https://sujal.hashnode.dev/your-ai-agent-said-it-completed-the-task-did-it-actually?utm_source=hashnode&utm_medium=feed)

[Nov 8](https://hashnode.com/posts/your-ai-agent-said-it-completed-the-task-did-it-actually/6af04d5a354245b83e1d71c2)

### [Your AI Agent Has Logs. So Why Is It Still So Difficult to Debug?](https://sujal.hashnode.dev/your-ai-agent-has-logs-so-why-is-it-still-so-difficult-to-debug?utm_source=hashnode&utm_medium=feed)

Why logging every LLM call doesn't automatically give you observability, and what it actually takes to debug an AI agent in production. Your AI agent is behaving strangely. Yesterday, it answered ques

17 min read · [sujal.hashnode.dev](https://sujal.hashnode.dev/your-ai-agent-has-logs-so-why-is-it-still-so-difficult-to-debug?utm_source=hashnode&utm_medium=feed)

[Nov 8](https://hashnode.com/posts/your-ai-agent-has-logs-so-why-is-it-still-so-difficult-to-debug/6af04d2eda93d76c0843c09d)

### [The Hardest Part of Agentic AI Isn’t Building the Agent. It’s Teaching It to Improve.](https://sujal.hashnode.dev/the-hardest-part-of-agentic-ai-isn-t-building-the-agent-it-s-teaching-it-to-improve?utm_source=hashnode&utm_medium=feed)

internalWe talk a lot about making AI agents more autonomous—giving them tools, memory, planning loops, and the ability to execute tasks. But there’s a more interesting question: How does an agent act

11 min read · [sujal.hashnode.dev](https://sujal.hashnode.dev/the-hardest-part-of-agentic-ai-isn-t-building-the-agent-it-s-teaching-it-to-improve?utm_source=hashnode&utm_medium=feed)

[Aug 29](https://hashnode.com/posts/the-hardest-part-of-agentic-ai-isn-t-building-the-agent-it-s-teaching-it-to-improve/6a934728c0a21c7efcaccb5e)

### [Your AI Agent Works. Now Make It Production-Ready](https://sujal.hashnode.dev/your-ai-agent-works-now-make-it-production-ready?utm_source=hashnode&utm_medium=feed)

Building an AI agent today is surprisingly easy. Give an LLM a prompt. Add a few tools. Connect an MCP server. Let it call APIs. Add some memory. And suddenly, you have an "agent." The demo works. It

10 min read · [sujal.hashnode.dev](https://sujal.hashnode.dev/your-ai-agent-works-now-make-it-production-ready?utm_source=hashnode&utm_medium=feed)

[Aug 23](https://hashnode.com/posts/your-ai-agent-works-now-make-it-production-ready/6a8b3ea8f6b662ead29afeec)
