Most first agentic projects fail on setup, not on the model.

This hands-on session is about getting the unglamorous parts right: your workspace, your tooling, and the conventions your team agrees before anyone starts prompting. Then we look at why the architecture you settle on before you build is what decides whether you ship something real, or a demo that cannot survive Monday.

PART 1 — Workspace hygiene for agentic development (~75 min)

The agent as a teammate, not a chat box. Giving it a codebase it can actually navigate: structure, and a context file that states your conventions and constraints. Extending it deliberately with skills, plugins and MCP servers, and knowing when adding one is the wrong answer. Secrets hygiene. And working as a team: four people prompting against one repo will produce four different codebases by lunchtime unless you agree conventions first.

PART 2 — Why architecture comes before building (~60 min)

The 24-hour scope trap. Architecture as a thinking tool rather than paperwork, and what actually breaks when you skip it. We will walk through real failures from building this event's own systems: a guard that silently blocked legitimate sign-ups, a screen that went blank after saving correctly, and dry runs that caught problems before they reached 300 people. Then, hands-on, your team drafts its own project proposal.

PART 3 — Troubleshooting and questions (~30 min)

Environments break. This time is set aside on purpose, so bring whatever is not working.

HOW TO JOIN

This is a hybrid session. Join us in person at GRD College, Coimbatore, or watch the live stream from anywhere: https://www.youtube.com/live/5FUVcKjBtXU — please register here either way so we know who is coming.

WHO IT IS FOR

Open to everyone. You do not need prior experience with AI agents. If you are taking part in Tech for Good 2026, bring your team: you will leave with your project proposal drafted, ready to submit.

WHAT TO BRING

A laptop with a terminal you are comfortable opening, Node.js and git installed if you can, a Google account for Gemini API access, and a problem you care about.

SPEAKER

Thiruppathi Krishnan, Application Developer at Ransky Tech, Coimbatore. Builds software with AI as a pair programmer day to day: the tooling, habits and guardrails that make an agent genuinely useful on a real codebase rather than a novelty. His engineering background spans automotive software and software-defined vehicles, HMI, and embedded systems.

Full details and agenda: https://t4g.gdgcbe.com/workshop