Jeremy FritzhandPrepared for ONDC
AI-Native Ways of Working.
A hands-on 2–3 day program that teaches your teams to direct AI — from the terminal and agentic tools to a structured knowledge base, live builds, and an executive console. Non-technical by design.
Jeremy Fritzhand · AI-Native Ways of Working© Jeremy FritzhandNot a lecture on AI. A working apprenticeship in using it.
Your teams already know their work. This program gives them the tools and the judgement to hand the repetitive, document-heavy, and technical parts to an AI they direct and review — and to walk out having built something real for their own function.
Direct, don't code
Participants learn to brief, supervise, and review an AI agent — not framework syntax. The skill is the instruction and the check.
Built, not slideware
Every session ends in an artefact: a knowledge base, a live page, an SOP workbook, a console — using the exact methods behind this deck.
Designed and delivered by Jeremy Fritzhand — who built a full institutional knowledge base, SOP system, and public tools with these methods.
AI-Native Ways of WorkingWhat this isThe line between technical and non-technical just moved.
Agentic AI can now read your files, run real commands, and build working software while a person directs and checks it. Work that used to require an engineer — structuring documents, drafting procedures, standing up an internal tool — is now within reach of the person who owns the problem.
Wait on engineering
File a ticket, join a queue, explain the context, wait weeks for a first version.
Direct an agent
Describe the outcome, review each step, and iterate to a working result the same day.
Capability in-house
Every function can structure its knowledge, ship internal tools, and report itself — without adding headcount.
AI-Native Ways of WorkingWhy nowThree sessions. Fifty-thousand feet to hands on the keys.
| Session | Focus | Teams leave with |
|---|---|---|
| Day 1 · Morning | The 50,000-ft view — terminal, IDE, agentic development, MCP, Claude's models & effort, connecting email & Drive, safety | A clear mental model and their own AI workspace connected to their tools |
| Day 1 · Afternoon | Building a structured, AI-readable knowledge base from the documents you already have | The skeleton of a real knowledge base for one function |
| Day 2 | From setup to shipping — Git & GitHub, deep research, fan-out agents, a live case study, and function SOPs → workbook → console | A deployable build and a repeatable operating method |
| Day 3 · Optional | A hands-on build lab: each function applies the method to its own work, with support | A finished KB + SOP workbook + draft console for their function |
AI-Native Ways of WorkingAt a glanceFive principles that keep this safe and real.
Bring real work
Participants use their own documents and problems — not toy examples — so the output is usable on Monday.
Use what you pay for
Claude, Gemini, or ChatGPT — the workflow is the same. We teach the method, not one vendor.
Scoped & reviewed
Institutional accounts, access scoped to a folder, and a human who reviews before anything is accepted or sent.
AI drafts, people decide
Nothing the AI produces is trusted until the person who does the work has checked and corrected it.
Start with one function
We go deep on a single team's workflow so the pattern is proven before it scales.
A method, not a demo
Teams leave able to repeat this without us — the whole point of the program.
AI-Native Ways of WorkingHow we'll workThe 50,000-foot view of systems & tools.
What the terminal is, what an IDE is, what "agentic development" means, how Claude's models and effort settings work, what an MCP is, and how to connect your email and Google Drive — in plain language, for people who have never written code.
AI-Native Ways of WorkingDay 1 · AMWhat is the terminal?
The terminal is a plain text window where you type instructions and your computer carries them out. It is Terminal on macOS and Linux, PowerShell on Windows — same idea everywhere: type a command, press enter, something happens.
- Runs commands — one line does one job.
- Manages files — create, move, delete.
- Installs tools — pull in what a project needs.
You will not memorise these. The AI types them for you — this is only so the screen is not a mystery.
$ npm create # start a project
$ npm install # add the tools
$ git init # track your workTerminal, then IDE, then the AI does it for you.
You do not sit at the terminal typing commands by hand. An agentic IDE sits on top of it and runs the real commands on your machine for you.
The AI decides which commands to run, runs them, and shows you the result. You watch and approve — you never type them by hand.
AI-Native Ways of WorkingDAY 1 AM · SYSTEMS & TOOLSWhat is an IDE?
An IDE is the workspace where your code and content live and get edited — one window that holds all your project files. An agentic IDE goes further: it can act on that folder itself.
EDITOR vs AGENTIC IDE
A plain editor is a place you type into. An agentic IDE — Claude Code, or VS Code with an agent — is a place you give instructions to, and it does the typing.
AN AGENTIC IDE CAN
- Read your folder — understand what is already there.
- Write files — create and edit content directly.
- Run commands — install, build, and test on its own.
- Explain itself — tell you what it changed and why.
AI-Native Ways of WorkingDAY 1 AM · SYSTEMS & TOOLSWhat is agentic development?
You direct, the AI executes, and you review. This is human-in-the-loop: you approve every step and check the changes — you never blindly accept.
AI-Native Ways of WorkingDAY 1 AM · SYSTEMS & TOOLSYour toolkit at a glance.
There are a handful of moving parts, but they fit together in a simple picture. You do not need all of them — and the ones you pick are interchangeable.
Make the thing
- Claude Code / VS Code — the agentic workspace
- Node.js & Astro — the site toolchain
- Git & GitHub — track and store the work
Direct the work
- Claude — plans, writes, and generates
- Gemini · ChatGPT — interchangeable alternatives
- MCP connectors — email, Drive, and more
Put it live
- Vercel — deploy and host, worldwide
- Resend — contact forms and email
- Analytics · Search Console — see what happens
AI-Native Ways of WorkingDAY 1 AM · SYSTEMS & TOOLSThe models are interchangeable — pick your lane.
Each major model comes paired with an agentic IDE. Any one of these will do the job. Choose the one your teams already use.
Claude → Claude Code
The Claude model, driving Anthropic's agentic coding tool.
Gemini → Antigravity
The Gemini model, driving Google's agentic IDE.
ChatGPT → Codex
The GPT model, driving OpenAI's agentic coding tool.
The workflow is identical whichever you choose. Learn it once; the vendor is a detail.
AI-Native Ways of WorkingDAY 1 AM · SYSTEMS & TOOLSWhere this pays off in your day.
Drafting
First-pass documents and emails, ready to refine.
Summarising
Long email threads and PDFs, down to the point.
Building tools
Small internal tools and sites for your functions.
Organising
Scattered files structured into a knowledge base.
SOPs & reports
First drafts your teams edit, not start from blank.
Research
Fast analysis and comparison across sources.
AI-Native Ways of WorkingDAY 1 AM · SYSTEMS & TOOLSYou stay in the loop.
Speed is the point, but so is accountability. The AI produces the draft; a person owns the outcome.
AI drafts, you validate.
Review every change before you accept it. Never let output go out unread. Keep a named human accountable for anything that leaves your desk — a document, a decision, a line of code.
AI-Native Ways of WorkingDAY 1 AM · SYSTEMS & TOOLSPick the model for the job.
Claude comes in three tiers. They share the same skills; they differ in how deeply they reason and how fast they respond. Match the tier to the task in front of you.
Opus
The most capable tier. Reach for it on hard, ambiguous, high-stakes work — strategy, thorny analysis, careful review.
Sonnet
Strong reasoning at good speed. The right default for most everyday work — drafting, research, summaries.
Haiku
Quick and economical. Use it for simple, high-volume tasks — fast lookups, sorting, short replies.
AI-Native Ways of WorkingDay 1 AM · Claude, MCP & Your WorldDial how hard it thinks.
Beyond the model, you set the effort — how much the AI thinks before it answers. It is like a colleague: sometimes you say "think it through", sometimes "just answer".
More thinking costs a little more time. Spend it where the decision actually matters.
WHERE TO SPEND IT
- Turn it up for — hard reasoning, planning, and reviewing important work.
- Keep it low for — simple lookups, quick edits, and routine replies.
AI-Native Ways of WorkingDay 1 AM · Claude, MCP & Your WorldWhat is an MCP?
MCP — the Model Context Protocol — is an open standard that lets your AI securely connect to your own tools and data: email, Drive, GitHub, databases, calendars.
Think of MCP as a universal adapter between the AI and your systems — one standard, many connectors, no custom wiring per tool.
AI-Native Ways of WorkingDay 1 AM · Claude, MCP & Your WorldPut your inbox to work.
Connect Gmail as a connector and the AI can search, summarise, and draft from your inbox — while you stay in control of what actually goes out.
The AI acts only within the access you grant — no more, no less.
AI-Native Ways of WorkingDay 1 AM · Claude, MCP & Your WorldA living knowledge base.
Google Drive works the same way. The AI can read your folders, pull the facts you need, and write new files back — turning static documents into a knowledge base you can question.
A connected Drive is what makes a knowledge base live, not static. This sets up the afternoon.
AI-Native Ways of WorkingDay 1 AM · Claude, MCP & Your WorldAudit your systems — the knowledge is already there.
Before writing anything new, point the AI at the systems your team already fills every day. Auditing them turns scattered records into a structured, searchable knowledge base.
Relationships & pipeline
- Contacts & accounts — who you work with
- Deals & stages — what is in motion
- Notes & history — what was said and decided
The written record
- Decks, sheets, docs — the work product
- Policies & SOPs — how things are done
- Reports — what has already been measured
The decisions
- Threads — context and commitments
- Approvals — who signed off on what
- Requests — recurring asks worth a template
Connect read-only first, with permission. The AI extracts and structures; a person reviews what enters the knowledge base.
AI-Native Ways of WorkingDAY 1 · CONNECT YOUR WORLDThe safety precautions that make this responsible.
The moment an agent can read your files, your inbox, and your Drive, good habits stop being optional. Five we set from day one.
Scope it tight
Point the agent at one folder and grant only the access it needs — never the whole drive by default.
Approve actions
Keep permission prompts on for commands, sends, and deletes. Read what it is about to do before you say yes.
Mind what is sensitive
Keep personal, legal, and regulated data out of AI-readable folders until policy clears it. Use institutional accounts.
Verify the output
Treat every draft as wrong until checked. The AI sounds confident even when it is not — a human signs off.
Keep a safety net
Version everything with Git and back up before big changes, so any step can be undone.
You stay accountable
The agent does the work; a named person owns the result. That never transfers.
AI-Native Ways of WorkingDay 1 · AMBuild a structured knowledge base.
Take the documents a function already has — scattered across folders, inboxes, and drives — and turn them into a tidy, AI-readable knowledge base: the source of truth that every deck, report, and SOP is generated from.
AI-Native Ways of WorkingDay 1 · PMA knowledge base is not software.
It is your work, organised — a tidy folder an AI can read, plus a memory of how your function thinks. You are not building an app. You are putting what you already have in order.
"I'd have to learn to code."
You do not. You point an AI at your files in plain English, and it does the moving, sorting, and writing for you.
"I'll break something."
There is nothing to break. It is folders and plain text — a copy of your own documents. You can always start again.
AI-Native Ways of WorkingDAY 1 PM · KNOWLEDGE BASEThe six parts of a knowledge base.
The tools
A short, standard kit an AI can drive on your behalf.
Your data
The documents, sheets, and notes your function already owns.
A folder taxonomy
A clean, named structure so every file has an obvious home.
Summaries and an index
One plain-language file per folder, plus a single map of the whole thing.
A design system
Your colours, fonts, and voice — written down once and reused.
A memory
A record of how your function decides: its rules and preferences.
AI-Native Ways of WorkingDAY 1 PM · KNOWLEDGE BASECompile, structure, enrich.
The AI builds the structure; people run it. Your job is direction and judgement — not typing.
AI-Native Ways of WorkingDAY 1 PM · KNOWLEDGE BASEThe skeleton, not the cathedral.
Start with one function. A base with gaps and [to-verify] tags beats a perfect document that never ships.
AI-Native Ways of WorkingDAY 1 PM · KNOWLEDGE BASEWhat a knowledge base lets you make.
It is the source of truth everything else is generated from — one tidy base, many outputs.
Branded decks
Presentations already in your colours, fonts, and voice.
Magazine-style reports
Long-form documents that read as finished, not draft.
Explainer web pages
Simple pages that walk a reader through one topic.
Trackers and SOP workbooks
Sheets and standard-procedure books your team runs on.
Compiled reports
One report pulled together from many scattered sources.
Day-to-day drafting
Emails, briefs, and notes written from what you already know.
AI-Native Ways of WorkingDAY 1 PM · KNOWLEDGE BASEThe honest effort — and cost.
Be honest about the effort. A skeleton takes an hour; a base your team actually trusts takes months of real use — after that, it needs almost no upkeep.
The method costs almost nothing in software; the investment is people's time and a few paid AI seats.
AI-Native Ways of WorkingDAY 1 PM · KNOWLEDGE BASEYour knowledge base answers questions.
A finished base is not a filing cabinet — it is queryable. Ask it in plain language and it pulls the answer from your own records: contacts, org structure, history, numbers.
Ask it things like
- Who is our contact at a partner, and what did we last agree?
- List every team member and what they own.
- Draft an org chart and reporting lines from what is on file.
- Pull the contacts tied to this program into a sheet.
Real example: a full institutional knowledge base holds the team, partners, portfolio, finances, and history — and answers all of the above from the repository itself.
> Pull every partner contact from
08-partnerships into a table:
name, org, email, last touch.
> Draft the team org chart from
01-organization/team-roles.md.From setup to shipping.
The technical foundations — Node, Git, GitHub, a host — then the advanced moves: deep research, fanning out agents, a real case study built this way, and how each function turns interviews into SOPs, a workbook, and an executive console.
AI-Native Ways of WorkingDay 2Day 2 setup: the five accounts and tools.
A short install session gets everyone on the same footing.
Node.js
The local toolchain that runs projects on your computer.
Git
Version history — a save-game for your work.
GitHub
Stores your repo online and connects it to deploy.
Vercel / GitHub Pages
Free hosting, plus a custom domain of your own.
Agentic IDE
Claude Code or VS Code — where the AI works alongside you.
Every tool has a free tier, and the AI walks you through each install step — you do not need to know how any of it works.
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHTwo ways to run Claude Code.
You can run the AI coding agent on your own machine or entirely in the browser. Pick by comfort level, not by rule.
On your machine (VS Code / terminal)
- Full access — it can read and edit all your files.
- Runs locally — it executes commands on your computer.
- Best when you are building something real to keep.
On the web (claude.ai/code)
- Cloud workspace — spun up from a GitHub repo.
- No install — everything runs in the browser.
- Best when you are new to coding.
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHFrom commit to live site.
You push your work to GitHub. The host builds it and serves it worldwide — on a custom domain, over HTTPS, automatically. No servers to manage.
Source stays in GitHub; the built site is served by the host. One push, and the change is live in seconds — no servers to manage.
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHGoing deeper: where data lives.
Hosting shows your site. It does not remember anything. For apps that store users, orders, or records, you add a database — Firebase or Supabase.
A database remembers
Stores users, orders, records, and settings — and lets an app create, read, update, and delete them.
The host is not the database
Hosting shows your site and runs functions. The data lives in a separate service alongside it.
Firebase or Supabase
Firebase: a NoSQL store with realtime sync. Supabase: a SQL (PostgreSQL) database. Both add auth and storage.
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHDeep research: send the AI to read for you.
The AI runs many searches, reads dozens of sources, and returns one cited synthesis — for market, regulatory, competitive, and landscape questions.
Always open the citations and check them. Treat the synthesis as a strong first draft, not a verdict.
GOOD USES
- Scheme & policy landscapes — what exists and who qualifies.
- Competitor & market scans — the field, at a glance.
- Diligence — background on a partner or startup.
- Literature & precedent reviews — prior work, pulled together.
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHFan out agents (ultracode).
The advanced move: run many AI agents in parallel. Each reads one part of a problem. Others adversarially verify every finding before it is trusted.
Used for exhaustive research, audits, and migrations too big for one context. The India schemes guide was built this way — one agent extracts, another independently re-checks every number.
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHSkills: a task you teach Claude once.
A skill is a small folder of instructions — plus any templates or scripts — that teaches Claude how you do a specific task. Save it once, and Claude follows it every time, the same way.
WHAT A SKILL IS
- A named capability — "draft our board update", "clean a CRM export".
- Reusable — write the how-to once; share it across the team.
- Consistent — the same steps, tone, and format every time.
HOW TO CREATE ONE
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHgstack: a ready-made library of skills.
You don't have to write every skill yourself. gstack is an open-source pack of 23+ skills for Claude Code that turns it into a virtual engineering team — plan, build, review, test, and ship, each a command away.
WHAT YOU GET
- Roles on call — a CEO, designer, engineer, QA, and security reviewer.
- A sprint you can run — Think, Plan, Build, Review, Test, Ship, Reflect.
- Slash commands — /office-hours, /review, /qa, /ship, and 20 more.
- Open source (MIT) — free to use and adapt.
INSTALL IN 30 SECONDS
git clone --depth 1 \
https://github.com/garrytan/gstack \
~/.claude/skills/gstack
cd ~/.claude/skills/gstack && ./setupgithub.com/garrytan/gstack — then add a gstack section to your project's CLAUDE.md.
AI-Native Ways of WorkingDAY 2 · SETUP & RESEARCHA 107-page government PDF, rebuilt as a living site.
In June 2026 the Government of India published the "Playbook of Government Schemes and Initiatives for Startups" — 107 pages of dense policy. The content is useful. The format makes it impossible to navigate.
- The problem — no founder finds the right scheme by flipping through a PDF.
- The rebuild — the same content, reorganised the way a founder actually looks for money: by stage, by sector, by geography.
- The result — a searchable site that stays current and costs nothing to run.
AI-Native Ways of WorkingCase Study · India Schemes GuideOne PDF, now a navigable map of the funding landscape.
BUILT IN, NOT BOLTED ON
AI-Native Ways of WorkingCase Study · India Schemes GuideExtract, verify, build, deploy — the same four moves.
Fan-out plus adversarial verification in practice: many pages built in parallel, each one checked by an agent that did not write it.
AI-Native Ways of WorkingCase Study · India Schemes GuidePoint it at any body of documents. The engine does the rest.
The India map is not a one-off. The same pipeline takes any structured set of documents and produces the same kind of site — searchable, accessible, and free to host.
WHAT YOUR TEAMS COULD REBUILD
- Network policies — the rulebook made searchable, not scattered across PDFs.
- Seller & buyer onboarding — one guided site in place of a static handbook.
- A living FAQ / handbook — updated by rerunning the pipeline, not rewriting a document.
Live at fritzhand.github.io/startup-india-guide.
AI-Native Ways of WorkingCase Study · India Schemes GuideEvery function documents itself.
Each team turns what it already does into onboarding docs and SOPs. Knowledge survives staff transitions, and new joiners get productive fast.
This is a capability your functions build for themselves, not a compliance box they tick. The people who do the work own the record of how it is done.
AI-Native Ways of WorkingDay 2 · OperationaliseBuild SOPs by interviewing your team.
AI DRAFTS · OPERATORS VALIDATE
Nothing is trusted until the person who does the work has corrected it.
AI-Native Ways of WorkingDay 2 · OperationaliseEvery meeting becomes minutes and action items.
Record the call, and the AI does the rest: a clean transcript, structured minutes, and a tracked list of action items with owners — filed straight into the knowledge base.
Nothing is lost between meetings. Minutes feed the SOPs; action items feed the console — so the record and the follow-through live in one place.
AI-Native Ways of WorkingDAY 3 · OPERATEWhat goes in the reporting template.
CAPTURED FROM EACH PERSON
- Name & role — who owns this work.
- Key responsibility areas — the core scope.
- Daily / weekly / monthly tasks — the real cadence.
- Data & metrics tracked — what they measure.
- Documents maintained — the records they keep.
- Suggestions — what should improve.
Definition
Process Owner
Approver
Frequency
Trigger
Process Flow (numbered)
Step | Action | Resp. | Approver
Output / record
Related documentsDraft the workbook, review in Google Drive, iterate.
REPEAT UNTIL CORRECT
The document is the data — you correct the source, not the polished output.
AI-Native Ways of WorkingDay 2 · OperationaliseRoll it up to an executive console.
One owner, one page, fixed shape, every month.
AI-Native Ways of WorkingDay 2 · OperationaliseThe console in practice.
FROM A REAL FACILITY BUILD — ILLUSTRATIVE FOR ONDC
Quarterly, each function's console rolls up so leadership sees the whole organisation in one shape.
Documents before dashboards. A leadership dashboard is only as alive as the workspaces that feed it.
AI-Native Ways of WorkingDay 2 · OperationaliseEvery participant leaves with artefacts, not notes.
Each participant
- Their own AI workspace, connected to email & Drive
- The skeleton of a knowledge base for their function
- A set of copy-paste prompts for their recurring work
- A safety checklist for scoped, reviewed AI use
Each function
- A method to turn interviews into SOPs and a workbook
- A one-page monthly console design for leadership
- A live, deployed example to learn from
- A repeatable operating pattern it can run without us
With the optional Day 3 build lab, each function leaves with a finished v1 knowledge base, SOP workbook, and draft console — not just the method.
AI-Native Ways of WorkingDeliverablesBuilt to fit your calendar and your rooms.
2 or 3 days
A 2-day core, plus an optional Day 3 hands-on build lab per function.
Small & hands-on
Sized so every participant is at a keyboard and gets attention. Multiple cohorts by function.
In person or remote
Delivered on-site or live online, adapted to ONDC's context and examples.
A laptop & an AI seat
A paid AI subscription per participant and admin rights to install a few free tools.
Bring real material
Each team gathers its documents beforehand — the raw input for the knowledge base.
Office hours
Optional post-workshop clinics while teams put the method into practice.
AI-Native Ways of WorkingLogisticsA path from a room of learners to an operating model.
Learn & build one
- Every participant builds a real artefact
- One function goes end-to-end as the exemplar
- Safety & review habits set from day one
Prove it in a pilot
- 2–3 functions complete their knowledge base
- SOPs drafted from interviews and validated
- First monthly consoles produced
Scale by evidence
- Train-the-trainer across functions
- Consoles roll up to a leadership view
- The capability lives in-house
AI-Native Ways of WorkingRoadmapSomeone who built this the hard way first.
AI-Native Ways of WorkingAboutGive your teams the ability to build.
I'd welcome a conversation about tailoring this to ONDC's functions and calendar. The next step is a short scoping call to choose the exemplar function and set the dates.
Jeremy Fritzhand · Prepared for ONDC© Jeremy Fritzhand