AI-Native Ways of Working

A three-day, hands-on AI workshop for ONDC teams. Prepared by Jeremy Fritzhand.

Program Overview

Agentic AI can now read files, run real commands, conduct research, analyse information, and build working software — while a person directs, reviews, and approves the work. This program equips ONDC teams with the tools and judgement to delegate repetitive, document-heavy, analytical, and technical work to AI systems they actively supervise — applying it to their own functions and leaving with something practical, built on real material. It climbs three levels.

Day 1 · Understand

The 50,000-foot view

What these tools are, how they work, and how to use them safely.

Agentic AI, terminals & IDEs — in plain language
How Claude works: models, effort & thinking
What an MCP is — connecting to email & Drive
Safety, permissions, and a human in the loop
Tool-agnostic: Claude, Gemini, or ChatGPT

Leave withA clear mental model and a personal AI workspace connected to your tools.

Day 2 · Apply

Hands on the keys

Build something real, from a knowledge base to a live webpage.

An AI-readable knowledge base from your docs
The toolchain: Git, GitHub, hosting, Claude Code, VS Code
Deep research; fan out agents in parallel
Worked example: a 100+ page scheme PDF → a live, searchable site
The daily loop: draft, review, correct, iterate

Leave withA knowledge-base structure and a deployable build you made yourself.

Day 3 · Operate

An operating model

Turn AI into how functions run and how leaders see performance.

SOPs from interviews, transcripts & your KB
An AI-drafted Excel workbook, reviewed in Drive
Each function's metrics on a one-page console
Governance for safe, accountable adoption
A 30–60 day pilot; documents before dashboards

Leave withA repeatable operating model and an executive-console design.

The Multiplier · Start from open foundations

Start from what's proven, not from scratch

The fastest builds begin on an open-source foundation. Name the job, search GitHub and each tool's ecosystem, check it's active and permissively licensed (MIT/Apache), then fork it and make it yours — re-skinned, wired to your data, with a person reviewing each step.

Open source · MIT
site2deck — turns a design system into branded, offline decks. This proposal runs on it.

Day 2 worked example
Startup India Guide — a 100+ page scheme, made into a searchable site.

Jeremy Fritzhand · AI-Native Ways of Working Page 1 / 2

Jeremy FritzhandAI-Native Ways of Working

Proposal for ONDC

Suggested Pilot Outcomes

By the end, ONDC could have…

A focused pilot — two or three functions chosen for AI enablement
Knowledge bases — structured, AI-readable sources for those functions
Initial SOPs — generated from interviews and existing documents
Working builds — internal tools, webpages, or operational workbooks

An executive console — a one-page monthly view for leadership
An adoption roadmap — a practical 30–60 day plan
Governance — defined review and human-approval processes

Format & Logistics

Built around your real work

Duration: Two to three days, run as one continuous engagement.
Delivery: In person or remote.
Each brings: A laptop and a paid AI seat.
Material: Real ONDC documents, processes, and functional material — wherever possible, participants work with their own material, not generic exercises.

Next Step

Let's run it with your teams

A short call is the fastest way to scope the pilot — which two or three functions, in person or remote, and the dates that fit your calendar.

Jeremy Fritzhand

Advising founders and institutions on AI-native ways of working

fritzhand.com linkedin.com/in/fritzhand calendly.com/jfritzhand

Jeremy Fritzhand · AI-Native Ways of Working Page 2 / 2