Blog — AI Agents, MCP Tools & Build Logs by Amit Kumar

Latest

• A2A Protocol vs MCP: What Google's Agent2Agent Standard Actually Solves (and What It Doesn't)28-08-2026

  The A2A protocol lets agents built on different frameworks talk to each other. Here's what it does, how it differs from MCP, and the gaps it leaves for production.

• When to Use AI Agents (and When a Plain Script Beats Them)25-08-2026

  Most teams reach for an agent before they know the task. Here's the framework I use to decide when to use AI agents versus a workflow or a plain script, with a real $40K build that proved the point.

• AI Agent Observability: Stop Flying Blind in Production21-08-2026

  A single agent run can touch 23 tool calls and cost real money. If you can't answer 'what did it do on run #4821,' you don't have an agent, you have a hopeful script. The TRACE model fixes that.

• How to Build Your Own MCP Server (in About 90 Minutes)18-08-2026

  Tutorials stop at installing someone else's MCP server. Here is how to build your own MCP server from scratch in Python, with a real schema and the security steps the quickstarts skip.

• Context Engineering: The Discipline That Makes AI Agents Actually Reliable14-08-2026

  Context engineering is the practice of designing everything an AI agent sees on every turn. A practical guide to the layers that break, the Root Theorem, and a production playbook.

• AI Agent Failures: Model or Harness? How to Localize the Real Fault07-08-2026

  Your agent failed again. Was it the model or the harness? A July 2026 paper catalogs 41 failure modes and assigns each one to the interaction where it originates. Here is the 5-minute triage I use to localize agent failures and route the fix to the right layer.