AI & the Modern Enterprise
01 · Promise
AI & the Modern Enterprise
Cardinality.AI · Guest session
AI & the Modern Enterprise
Connecting People & Process
Kirubaharan M S
Senior Vice President & India Site Lead, Cardinality.AI
02 · Promise
Three things you’ll understand
A promise
By the end of this talk, you’ll understand three things.
01
How LLMs are different
Why this wave isn't just another tool.
02
The Great Compression
Why the next 25 years may hold 2,000 years of change.
03
How it lands in an enterprise
Where agents act, and what humans must own.
03 · Promise
The real question
The question
“What can AI do?”
The better question
“How do we make AI work for the enterprise?”
04 · Context
The three divides
200 years of technology
What has technology been erasing?
Steam · 1800s
The Muscle Divide
Machines lifted what bodies couldn’t.
- · Strength no longer set your worth
- · Industry erased the Muscle Divide
Internet · 1900s
The Access Divide
Anyone could reach anything.
- · The bank, the library, the office came to you
- · Distance stopped deciding your future
AI · Now
The Knowledge Divide
Expertise on tap, for everyone.
- · Reasoning becomes a utility
- · No degree needed to build
First we freed the body. Then the reach. Now the mind.
05 · Context
Back to 1870
Today
2026
Let’s go back in time.
Hold on…
06 · Context
Fast forward to 1900
Same family
1870
Now, fast forward.
07 · Context
What 30 years added up to
1870 → 1900
What did 30 years of change add up to?
2×
Labour productivity
4×
GDP
6×
Real manufacturing
7×
Stock market
We are in 1870 again. The rebuild is about to begin.
08 · Context
The invention vs the rebuild
The lag
Electricity arrived. Why didn’t factories change overnight?
- Swapped the steam engine for a motor
- Kept everything else the same
- Gains came only when factories were rebuilt
- That took about two decades
The invention wasn’t the revolution. The rebuild was.
09 · How LLMs differ
AI is 70. What’s new?
What’s actually new
AI is 70 years old. So what’s new now?
1956
Dartmouth coins “AI”
1997
Deep Blue beats Kasparov
2012
Deep learning takes off
2017
Transformers
2022
ChatGPT
01
Agents decide
on their own.
02
Agents invent
what wasn’t there.
03
Agents keep
improving themselves.
10 · How LLMs differ
It aims at judgment
An objection worth naming
- Steam lifted loads, but never chose what to lift
- The internet delivered facts, but left thinking to us
This technology aims at judgment itself.
A deeper wave is a reason to design, not to freeze.
11 · Great Compression
The Great Compression
The Great Compression
How much change can fit into 25 years?
The next 25 years may contain as much change as the previous 2,000.
12 · Great Compression
The steepest curve
The steepest curve
How long does it take a new idea to reach everyone?
13 · Great Compression
The growth of AI
The growth of AI
Is this hype, or is the money real?
Investment
$415B
$820B
Big-5 hyperscaler capex, 2025 → 2026
≈ 2.5% of US GDP
Valuation
$683B
$2–3T
Anthropic + OpenAI combined, Sept ’25 → today (est.)
Revenue up 100× in two years
Tokens
480T
3.2 quadrillion
Google monthly tokens processed, → May ’26
From almost nothing in 2024
14 · Great Compression
Fast forward to 2050
Remember
1900
30 years turned a farmhand into a city worker.
15 · Enterprise
Adopt or redesign?
Phase two of enterprise AI
Adopt AI, or redesign around it?
- Phase one was adoption: copilots, pilots, governance
- Next: redesign how the enterprise creates value
- Rethink operating models, not just tools
Technology enables. Redesign transforms.
16 · Enterprise
Rules or judgment?
Where humans matter
Rules, or judgment?
- Simple decisions → rules, thresholds, automation
- Judgment: incomplete info, conflicting goals
- Judgment: when the normal rule doesn’t fit
Design how AI reasoning and human judgment work together.
17 · Enterprise
Is data enough?
Enterprise intelligence
Is giving AI your data enough?
- No. AI also needs context and relationships
- Start from the outcome, reason backward
- Build shared foundations: data, semantics, governance
Let the work strengthen the foundation.
18 · Enterprise
The AI-native organization
AI-native organizations
What makes an organization AI-native?
- AI is a participant in knowledge work, not a tool
- Rethink decisions, collaboration, knowledge flow
- Measure success differently
AI-native is designed, not adopted.
19 · Enterprise
Tool or agent?
Tool or agent?
What’s the difference between a tool and an agent?
A tool completes a step. An agent chooses steps toward a goal.
An agent is a tool that needs a boundary.
20 · Enterprise
The first decision
A scholarship case
The names don’t match. Approve, reject or escalate?
Ask first: which evidence would change the decision?
The fastest answer may be the one that forgot to ask a question.
21 · Enterprise
The approval trap
The approval trap
Three hundred cases. One reviewer. One Approve button.
Real review needs evidence, policy, time and authority to challenge.
A human in the loop needs more than a button in the loop.
22 · Enterprise
Earned autonomy
Earned autonomy
Which actions can we verify and undo?
Start with extraction, checks, routing and drafts. Expand as trust is earned.
Give AI the work whose mistakes we can find and repair.
23 · Enterprise
When should the agent decide?
When should the agent decide?
Same AI. Opposite answers.
Radiology · let it learn
One doctor: 60 years × 20 reports a day
0
scans in a lifetime
AI: millions of scans overnight, learning from confirmed outcomes.
A finite life vs an infinite library. Here the agent can outgrow the expert.
Share market · be careful
- Every agent learns the same data.
- They all see the same signal.
- They all sell together.
Here, speed is the risk, not the reward.
Monoculture of intelligence → Algorithmic stampede
When everyone’s agent thinks alike, the market falls alike.
24 · Enterprise
The bottleneck moves
The bottleneck moves
The extraction takes seconds. Why does the student still wait days?
A faster task can simply move the queue downstream.
Redesign the workflow, not only the automated step.
25 · Enterprise
ROI and certainty
ROI & certainty
Why don’t enterprises just switch AI on?
- They buy outcomes, not models: ROI, certainty, audit trails.
- Start where value is measurable and mistakes are repairable.
- Measure end-to-end impact, not demo speed.
Enterprises don’t pay for intelligence. They pay for certainty.
26 · Enterprise
Business mapping
Enterprise business mapping
How do you map a business for AI?
e.g. Unit test writing · Software Engineer · test frameworks, edge cases · TestGen Agent · 4.0 hrs → 0.5 hrs (88%)
0
departments
0
activities
0
micro-skills
0
agents
0 / 0
fully / partly automatable
Est. annual savings
$0
Human $45/hr · Agent $5/hr · 80% effectiveness
Map the work before you buy the AI.
27 · Enterprise
Clip: why pilots fail
Clip 1 · The 95% problem
Why most AI pilots never reach production
▶ Click to play the clip
Pilots impress. Production pays.
28 · Enterprise
Clip: the last mile
Clip 2 · The last mile
Why enterprise AI breaks at scale
▶ Click to play the clip
In mission-critical work, 95% right is still wrong.
29 · Enterprise
Where’s the value, India?
Where’s the value, India?
Enormous sums are going into AI. What’s in it for India?
UPI-scale rails
Digital public infrastructure for a billion people
IndiaAI Mission
Shared GPU compute for startups and researchers
Data centres
Buildout across Mumbai, Chennai, Hyderabad
Talent
One of the world’s largest developer bases
So what is actually scarce?
30 · Close
Shameless marketing
I was told not to do this in a student talk…
…so I made it a slide.
31 · Close
The questions I couldn’t stop asking
The questions I couldn’t stop asking
Will AI take my job?
Should my child still bother learning to code?
Can I trust anything I see on a screen anymore?
If a machine can care for my parents, what am I for?
For ten thousand years, there was not enough intelligence to go around.
That fact is dying.
32 · Close
The Age of Lasts
The Age of Lasts · Artificial Abundance
There was never enough intelligence to go around. Until now.
Written. In print. On Amazon by late October 2026.
Which room of your life is changing fastest right now?
For deeper answers, buy the book. For shallower ones, ask ChatGPT. 😄
33 · Close
Expectation vs reality
Meanwhile, in the boardroom
Expectation vs reality
What the board expects
Install AI Monday. Profit Tuesday. 🚀
What actually happens
Month 6: still cleaning the Excel sheet. 📊😩
34 · Close
Things don’t change overnight
A realistic expectation
Things don’t change overnight.
1900Only horses
1920Horses and cars, side by side
1950Only automobiles
People were trained to ride horses, not drive cars. Skills take time.
Owners had invested in horses and stables. Old investments slow change.
Yet the road changed completely. Not if, but when. And who is ready?
35 · Close
Ask the horse
Hold the reins. AI is the horse. You are the rider.
36 · Close
Three quotes
Three things to carry out of this room
“The technology that scares us most eventually becomes invisible.”
“The greatest skill of the AI age is deciding what deserves a human.”
“AI owns the process. Humans own the purpose.”
37 · Close
A question for life
Every generation gets one great tool.
Yours can think.
A question to carry for the rest of your life
What will you build that only a human could dream of?
Thank you · Kirubaharan M S