How a 12-year-old company rebuilt itself AI-native — and the method any company can copy.
press → to begin
Cue: No bio. Open straight into the paranoia. Tell them what they'll leave with: the method + a one-page diagnostic.
Your host
Anuj Garg
Builder, GenAI
Cue: One breath. "I'm Anuj, I build GenAI." Then move — no bio.
01
Section 1 · Where this started
One paranoia rewired everything.
Not a roadmap. A fear.
Cue: Chapter card. Pause a beat, then go to the $401M slide.
01 · PARANOIA
The question that started it
Two builders in a garage could rebuild any of us — AI-first, from scratch.
So we asked one question: if someone beats us, is it on cost — or on something we'd never see coming? We decided to become that company first.
Vikram Chopra · Founder & CEO
"We are way, way behind."
Cue: The hook is the "two builders in a garage" fear. The rest of the founder's words — "do or die, I'm paranoid" — tell as spoken story, don't put it on the slide.
01 · REFLECTION
Your turn · 90 seconds · on paper
If two people rebuilt your company AI-first tomorrow — what would they do that you'd never see coming?
Write it down. Fold it. We come back to it at the end.
Cue: Actually wait the 90 seconds. Don't collect — it's their private blind spot. This bookends the close.
02
Section 2 · A diagnostic
Three levels of AI.
Where are you — really?
Cue: Frame this as a mirror. They'll place themselves in the next slide.
02 · THE LADDER
Three levels of AI in a company
1 Bolt-on — sprinkle AI on old workflows. Most of the market.
2 Embed — industrialize AI into existing workflows. Where most leaders think they are.
3Rebuild— re-architect around AI. Ask what should stop existing. The real prize.
Most companies are Level 1 — convinced they're at Level 3. You can't reach Level 3 by adding AI to the company you have.
Cue: Show of hands per level — calibration, no wrong answers. Then the "calling it Level 3" line. Ask: "what would you build with no legacy?"
03
Section 3 · The method
Build rails, not trains.
Trains are use cases. Rails are what anyone runs a use case on.
Cue: The centerpiece — this is your IP. Give it the most time. Repeat the four words.
03 · THE TRAP
Why most AI stalls
You're building trains.
Someone sees a use case, builds a bot. Then another. Six months later: six tools, six teams, nothing talks. A yard full of trains and no tracks.
A list of use cases isn't a strategy. It's a shopping list.
Cue: The enemy. Be a little provocative — most of the room is guilty of exactly this.
03 · RAILS
What we did instead
Build the rails first.
Agent brain
Airo
Build once, deploy across chat, voice, WhatsApp, CRM.
Workflow + tools
Weave
Chain any agent to internal systems, no code deploy.
Support automation
Axis
Routes, resolves, and remembers across every channel.
Docs + calls
Doc & Audio Intelligence
Reads any document and turns every call into structured insight.
Vision
Visio
Standardizes and scores every car image at scale.
Memory
Chronicle
One memory of every customer any agent can read.
A new use case took 15 minutes of engineering, not six months.
Cue: Don't read every box. The point: these are what EVERY agent needs. Build once, reuse everywhere — that's the compounding.
03 · THE REVEAL
Why we could move so fast
The rails took years. The agents took days.
When GenAI arrived, we didn't start from zero — the hard part (our data, our models, our pipelines) was already built. The moment the models got good, anyone could build on top in days.
The winners won't be fastest to GenAI. They'll be the ones who already laid rails.
Cue: Your mid-talk "one more thing." Drop your voice. Reframes the whole story: not speed of adoption — infrastructure that was already there.
03 · THE PAYOFF
My corner of it
45+ agents live. Built in days.
I run our conversational AI. We put the buyer journey on WhatsApp and bookings roughly doubled — because it rode rails that already knew the customer, the catalog, the documents.
Most were built by people who aren't engineers.
Cue: Keep numbers directional. The punch is the last line — non-engineers. It sets up Section 4.
03 · EXERCISE A
Exercise · 6 minutes · canvas side A
Trains vs Rails audit
1
List everything you built or bought in AI this year.
2
Tag each: T for train (one use case) or R for rail (reusable infra anyone builds on). Count.
3
Name the one rail that, if it existed, would let 5 of your trains ship in a week.
Most of you will find your rail count is near zero. That's the trap — and the way out.
Cue: 3 min solo, 2 min pair-share, 1 min harvest. Get one person to read their "one rail" aloud.
04
Section 4 · The org
Builder is the only role left.
When the rails exist — who builds?
Cue: This is the boldest chapter. Own it.
04 · FLATLAND
The org redesign
We dropped titles.
We replaced every title with one role: Builder.
Hierarchy was never a design choice. It was a coordination tax — humans passing context around. AI holds the context now. The layers became optional.
Vikram Chopra · Founder & CEO (public on X)
"Builder is the only role left in a company. AI just mass-produced that capability for everyone."
Cue: "Headcount is not capability." Pay off the horses/cars plant here.
04 · THE PROOF
The signal it's real
I'm a civil engineer. And an MBA. Not a software engineer.
This week I shipped two features to production.
When the rails are there, your degree stops mattering. Everyone can build. That's Flatland — and I'm not the exception. Across our company, people who've never written code ship every week.
Cue: Two archetypes: the PM (talks→builds) and the site manager (zero code→shipped). Makes the abstract claim human.
05
Section 5 · What nobody tells you
Hold the J-curve.
Every agent is worse than a human — at first.
Cue: Shift tone. Drop the bravado. This honesty is what they remember you for.
05 · MY SCAR
Now let me tell you what the case studies never do
50%
My AI voice agent launched at half a human's efficiency.
Show a board that number and they'll kill it. Most companies do — they quit at week two.
Cue: Your scar. Say it slowly. Callback: "Remember 'we're way behind'? I felt that personally." Then the grind: we pulled apart every part and listened to the calls.
05 · THE BREAKTHROUGH
The fix wasn't a better copy
Don't build a better human. Build what a human can't be.
We kept the human's workflow but played to the AI's strengths — call the lead the second it lands, ten thousand at once, 9pm, never tired. Then it didn't catch up. It passed.
Cue: THE line they quote you on. Two seconds of silence before you explain it.
05 · THE DISCIPLINE
How you hold the line
Match 90–95%. Then scale. Hold the line.
The model is the easy part. The plumbing is the moat.
An agent went 0-for-15 on live calls — yes 15 times, zero went through. The model was perfect; a database was down for an hour. We only caught it by listening to the calls, not the dashboard.
Don't automate the spaghetti.
A company we studied shipped a brilliant copilot in two weeks. Zero adoption. If your process is broken, AI just runs it broken — faster.
Cue: Land the formula: transformation = rails + patience + the guts to rewire the org.
06 · YOUR MOVE
Section 6 · Unfold your paper
Can you survive your blind spot by adding features — or does it need new rails?
Monday, measure one number: the hours from "I know what to build" to "someone starts building it." That gap is your transformation debt.
And ask yourself: is my work 50% different than 100 days ago? Am I shipping — not delegating?
Cue: Bring them back to the blind spot they wrote at the start. Have them write the Build Proximity number now.
RememberRails, not trains.
RememberBuilder is the only role.
RememberHold the J-curve.
Don't bolt AI onto the company you have. Build the company AI makes possible.
Build the rails. Everyone becomes a builder.
Anuj Garg · Builder, GenAI · if you're working through your rails, let's talk.
Cue: Three takeaways slowly, like a closing chord. Soft advisory invite. Then the final line — stop. Don't add anything after it.