VARAHI TECHNOLOGIES · PUNE
AI-FIRST BUILDER SERIES · VOL. 01
25.04.2026 · 09:00 IST
The AI-First Builder Series / Session 01

Anatomy
of an
Agent.[001]

Hands-on · Limited to 15 builders

A 3-hour working session for founders and engineers building AI-first products. We open two real agentic codebases on screen and read them together — architecture, tool contracts, permission models, sub-agent orchestration — and walk out with a scaffold to build our own.

Date Sat 25 Apr 2026
Time 09:00 → 12:00 IST · 3 hours
Format Hybrid 15 in-person · open livestream
Fee No fee Application based
N 18.538° · E 73.682°
Venue
Varahi Technologies
305, Delta Square, Bhukum, Pirangut, Pune
DELTA SQUARE · 305
§ 01
Why a reading session, not another workshop.

Everyone is shipping "agents." Very few builders have actually read how a production-grade agentic system is wired — the tool loop, the permission model, how MCP servers get negotiated, how plan mode differs from execution mode, how sub-agents are spawned.

That gap is where most AI-first products get stuck: prototypes that look magical but break the moment you ask them to do real, multi-step, stateful work.

The fastest way to learn how to build serious agentic systems is to read the serious ones that already exist — and then build.

§ 02
Two codebases, one question: what does good look like?
Reference A · ~1,900 files

Claude Code CLI

TypeScript · Bun · React + Ink · MCP · LSP
  • Tool system — ~40 tools: BashTool, FileEditTool, AgentTool
  • Core query engine — streaming, tool loops, thinking mode, retries
  • Permission hooks, plan mode, bypass & auto modes
  • MCP service layer and coordinator for sub-agent orchestration
  • Feature-flag driven dead-code elimination at build time
  • Bridge system for IDE integration (VS Code, JetBrains)
github.com/codeaashu/claude-code
Reference B · MIT · monorepo

pi-mono

TypeScript · 7 packages · vLLM · OpenAI/Anthropic/Google
  • pi-ai — unified multi-provider LLM API
  • pi-agent-core — agent runtime, tool calling, state
  • pi-coding-agent — interactive coding CLI
  • pi-mom — Slack bot that delegates to the agent
  • pi-tui + pi-web-ui — agent-ready UI primitives
  • pi-pods — vLLM deployment on GPU pods
github.com/badlogic/pi-mono
§ 03
The three-hour working agenda.
09:00 · 09:30
Block 01 · 30 min

Informal networking over chai

Arrive, grab a cup of chai, meet the room. No icebreakers, no forced intros — just 30 minutes to put faces to handles before we dive into the code.

09:30 · 11:30
💻
Block 02 · 2 hours · main event

Collaborative code walkthrough + brainstorm

Two hours on Claude Code CLI and pi-mono, side by side. Architecture, tool contracts, permission models, streaming, sub-agent orchestration. Participants who've already studied the repos lead segments. Everyone else brings questions, pushback and their own building context. The goal: extract the patterns both systems converge on — and where they diverge, and why.

11:30 · 12:00
🚀
Block 03 · 30 min · demo + close

Founder demo + open conversation

A short live demo from one of the participating founders showing what they're building in agentic AI today. Walk out having seen a real AI-first product in action — not just read about one. Followed by open Q&A and an invite to the follow-up study group.

§ 04
Who this room is for.
You'll fit right in if
  • You're a founder, CTO or senior engineer building AI products.
  • You ship code, not slideware — hands-on is the baseline.
  • You're past the prompt-engineering phase and want real architecture.
  • You think in systems: tool contracts, state, permissions, streaming.
Skip this one if
  • You want a beginner "what is an LLM" intro.
  • You're looking for a no-code tool walkthrough.
  • Reading TypeScript on screen isn't your thing.
  • You're expecting a passive webinar — this is a working session.
§ 05
What you walk away with.
01

A working mental model of how a real agentic coding system is structured.

02

A side-by-side comparison of two production-class architectures, trade-offs called out.

03

A starter scaffold to begin building your own agent loop — a foundation, not a toy.

04

A live look at what a fellow founder is shipping in agentic AI right now.

05

A circle of serious AI builders in Pune — and an open invite to future sessions.

§ 06
Come prepared.
01 · SETUP
Laptop, Node 20+, Git

You'll be coding along in the final stretch, not just watching.

02 · REPOS
Both codebases cloned

Clone links are shared the day after registration so you land ready.

03 · API KEY
One LLM provider key

Anthropic, OpenAI or any provider of your choice — for the build-out.

04 · OPTIONAL
30 minutes of README skimming

Highly recommended. You'll get 2× more out of the deep-reads.

§ 07
How to join.
© 2026 Varahi Technologies Pvt. Ltd. Set in Fraunces · Inter Tight · JetBrains Mono ANATOMY OF AN AGENT · V04 · APR 2026