anantgupta.xyz
→ · space · clicker
iamanantgupta
Anant Gupta · August 2026

What one person
can ship now.

A field report from a year of building with Claude Code.

Every slide you're about to see
was built by the thing it's about.

I didn't design this deck. I described it.

The last twelve months, in numbers
150+
projects on this laptop
50+
custom tools it built for itself
6
production servers it runs
2,900+
working sessions, all remembered

One laptop. Real work that carries revenue, not demos.

Quick intro

I'm Anant.

I'm a founder, not an AI researcher. I run B2B SaaS products: WhatsApp marketing, lead generation, email infrastructure.

For the last year, Claude Code has quietly been my co-founder. Everything you're about to see is real and in production today.

The short backstory

In this space since 2019.

2019
Started in college, at 19. My first product was an Instagram email scraper. It got featured in one of the biggest cold-email courses on the planet, then Instagram's lawyers asked me to shut it down.
ALSO 2019
Already using AI, before it was popular: my chatbots ran on Google Dialogflow. Customers kept asking how I replied so fast. I've been practicing this ever since.
SINCE THEN
B2B SaaS used in 90+ countries, built with my brother. Bootstrapped, paid for by the people using it, zero outside money.
LAST YEAR
Claude Code entered the workflow. That's the year this talk is about.

The ideology behind all of it.

01
Buy back your time.
Every hour a machine works is an hour you get back. Automation isn't about output. It's about ownership of your day.
02
Fix your health.
Some of those bought-back hours went into my body. Later you'll see the app I built to make that stick.
03
Create leverage.
One person plus a tireless agent compounds like a team. Claude Code is the biggest lever I've ever held.

My promise for the next 45 minutes.

SEE
Real things it shipped.
Ads, dashboards, video, whole products. All from my actual business, no toy demos.
WATCH
Live demos, not slides.
Eyes on my screen, hands on my browser. At one point it will text my phone. Stay for that.
LEAVE
With the playbook.
The exact three moves to start tonight, and the one mistake that makes people quit in week one.
01

The fundamentals

Claude vs Claude Code, in plain words.

4 min be at 0:10
Claude, in a browser tab
How do I A/B test my pricing page?
Great question! Here are seven steps you could follow. First, choose a testing framework…
Claude Code, in my terminal
claude — ~/whatchimp
> A/B test the pricing page
⏺ Read pricing/index.tsx
⏺ Edit variant B · headline + CTA
⏺ Test ✓ 42 passed
⏺ Deploy ✓ live
Split running. Report in 7 days.

Same brain. One can only talk. The other can act.

It has hands.

Terminal Files Git Browser APIs Servers

Claude Code is a terminal app: you type plain English, and it works where the work actually lives. So it can finish things, not just suggest them.

The unit of work changed
“Write me a function.”
“Build me the feature.”
“Run this part of my business.”
02

Shipped

Real things my business runs on. With receipts.

14 min be at 0:24

Most of what follows
is from one product.

WhatChimp: WhatsApp marketing for businesses. My main lab for paid ads, funnels, docs, support, dashboards.

Marketing

It ships my Meta ads.

Designs the creative, writes the copy, uploads the carousel, publishes the campaign. Straight through the Ads API.

The static on the right? It drew it in code. No designer, no Figma.

A finished WhatChimp comparison ad creative, drawn entirely in code
Marketing

It renders video.

Drawn frame by frame in HTML, encoded with ffmpeg, scored and voiced through APIs. No editor was ever opened.

These are frames from two finished ads for WhatChimp.

Frame from the race ad video
Frame from the inbox ad video
PLAY the race ad, 30 sec, with sound ↗
Marketing

It debugged my funnel.

Ad attribution was silently broken: a caching layer was wiping the tracking data. It found the bug, wrote a WordPress plugin to fix it, and wired server-side conversions back to Meta.

The kind of bug that quietly wastes ad spend for months.

Operations

It builds the dashboards I check every morning.

Live revenue, spend and LTV, pulled from the sources of truth and deployed on my own servers.

metrics.anantgupta.xyz
The live WhatChimp revenue dashboard
OPEN metrics.anantgupta.xyz ↗
Operations

It built my team's command center.

Renewals at risk, onboarding queues, win-backs, the team inbox. My whole support operation runs from a panel Claude Code built end to end.

success.whatchimp.com
The WhatChimp admin command center
OPEN success.whatchimp.com ↗
Support

It writes my help center. In six languages.

Every article drafted, illustrated with real screenshots, and published through an API connection it built for itself. English plus five translations, kept in sync.

help.whatchimp.com
The WhatChimp help center
OPEN help.whatchimp.com ↗
Personal · the fix-your-health part

It built my nutrition coach.

Remember the ideology? This is it in action. It syncs with my real coach's spreadsheet, reads my meal photos and estimates macros, and answers like a sharp friend: sixty words, no lectures.

OPEN track-health.anantgupta.xyz ↗

…and 140 more.

SendMailsharraka-smtpvideo-call-saasgmb-scrapercarousel-makerlead-generation-saaslinkedin-automationads-library-watcherdisposable-email-saasappointment-systemdomainbuyhetzner-deployerinstagram-dm-saasfb-comment-cleanerclaude-operatorgrowthtoolkit-mcpclosechimpindie-kit-proembedding-aiflow-buildercamera-tutorialpeople-portalgitadaily-offloadclone-websiteapify-actorwhatchimp-supporthr-portalallinonechatweb-scraperig-learneryt-learnerloom-watchernews-clibrand-fetchclaude-whatsapp

Every tile is a real folder on this laptop.

03

Giving it a body

Where it stopped being a tool and became a colleague.

9 min be at 0:33
Sense 01

Eyes.

One command and it captures my screen. It sees the bug, the dialog, the thing I can't describe.

$ claude-eye active
LIVE it reads my screen while you watch
Sense 02

Hands.

It drives my real Chrome: my logins, my tabs, my ad accounts. It even built the Chrome extension it uses to do this.

$ browser click "Publish"
LIVE drive the browser
Sense 03

It watches video.

Drop a YouTube link. It pulls the transcript and the frames, learns the technique, then builds a tool from what it learned.

$ yt-watch youtu.be/…
Sense 04

It has my number.

When it's blocked, it texts me on WhatsApp. I answer with a voice note from anywhere, and it keeps going.

harraka-server · CedarDeploy finished ✓ 3 services green. Cert auto-renews Sep 2.
11:42 PM
whatchimp · AtlasBlocked: need the staging Stripe key to run the checkout test. Reply when free.
11:47 PM
0:07
11:52 PM ✓✓
LIVE have it text my phone, now
Sense 05

It remembers.

2,900 sessions, indexed and searchable. Every new prompt automatically recalls the relevant past: decisions, bugs, preferences. “Did we solve this before?” always has an answer.

Sense 06

It delegates.

Big job? It opens more terminals and runs a team of itself: one researching, one building, one testing. Then it collects the results.

$ term-spawn · term-task
Sense 07

It compounds.

A hook catches every correction I make and writes it into a charter that every future session must obey. I never give the same feedback twice.

My design taste file is now 2,000 lines of my own corrections. That's why nothing it ships looks like AI slop.

04

The toolkit

Three words you'll keep hearing: skills, MCPs, APIs.

4 min be at 0:37

Skills: teach it a playbook once.

/taste-lock
My entire design standard. Loads before any UI work so nothing ships generic.
/motion-ads
The whole video ad pipeline: frames, encoding, music, narration.
/ship-it
Lint, test, commit, deploy. One word instead of five commands.

A skill is a file of instructions it loads on demand. Write the workflow once, invoke it forever.

MCPs: plug in new hands.

playwright
Drives and tests any website in a real browser.
scraper
Reads any webpage or doc site as clean text.
docs
Publishes straight into my help center.
hetzner
Provisions and manages my servers.

An MCP is a connector. Anything with an API can become a tool it holds.

APIs: you only need to know
they exist.

Every app is a thin skin over an API. If a website can do it, Claude Code can do it for you: on schedule, at scale, while you sleep. That is where the possibilities open up.

05

What I believe now

4 min be at 0:41

Watching isn't learning.
Artifacts are.

Every video, doc, or idea has to leave behind a tool, a rule, or a memory. Otherwise it evaporates by tomorrow.

Let AI think once.
Let code repeat it.

How most people use AI
Ask the AI to do the task, every single time.
It re-thinks the whole job from scratch.
Tokens burn fast. Results drift. Speed caps out.
The deterministic move
AI does the job once, while you watch.
You spot the repeatable pattern in it.
It writes code for that pattern. The code runs it forever: no thinking, no drift, near-zero cost.

Non-deterministic for the new. Deterministic for the repeatable. That's the whole trick.

Do it three times?
Make it a tool.

Code is memory that doesn't fade. My laptop has fifty commands that exist because I got tired of repeating myself.

Taste is
the bottleneck.

The model is the same for everyone. Output quality is set by the standards you refuse to lower.

Everyone has the same model.
Nobody has the same system.

The memory, the tools, the rules: that's the compounding. That's the moat.

06

Start tonight

4 min close at 0:45
As promised: why people quit in week one

They treat it like a chatbot.

They ask it questions, get answers, and shrug. It only clicks when you give it work it can own end to end: a task with files, a result, a finish line.

Three moves, in order
01
Pick one real task.
Not a toy. Something that annoys you weekly: a report, a cleanup, a page. Real stakes teach you what it can do.
02
Write down your context.
What you know, how you like things done, what it should never do. Context is the difference between generic and yours.
03
Make corrections permanent.
Every time you fix its work, save the rule. Six months of that is a system nobody can copy.
One last picture before you go

Each dot is 3.2 million people.

Never used AI · 6.8B (84%) Free chatbot · 1.3B (16%) Pays for AI · ~20M (0.3%) Uses agents · ~2‑5M (0.04%)

Tonight you join the smallest group on this chart.

Tonight

Install it. Give it one task.

$ npm install -g @anthropic-ai/claude-code

Mac, Windows, Linux. Or in the browser at claude.ai/code

CHAT type the one task you'll give it tonight

The question isn't
what AI can do.
It's what you'll let it own.

Thank you. Anant