Ashish Muralidharan
IISc · March 6, 2026
A question for this room
How many of you feel anxious about your future because of AI?
Live Results
0 responses
Very anxious
0%
Somewhat
0%
Not really
0%
Excited
0%
Ashish Muralidharan
Why am I trying to answer this for you?
A combination of three threads in my career gives me a unique lens on this question.
Productivity Research
I build Excel and Excel Agent at Microsoft. Office products have been the backbone of productivity across industries, functions, and profiles. Understanding how people are productive at work, and how AI can enhance it, has been a core research objective for the last two years.
Career Building
As prep head at IIM Calcutta, I mentored 1,000+ students across three batches to find the right career fit and break into their target industries. I continue to mentor students and professionals through career coaching and corporate learning programs.
AI Coaching
Coached 100+ product managers in India through career transitions. Now coaching 100+ trainers building AI PM programs across the US. This gives a practitioner's lens on how AI is reshaping career paths.
Productivity research + career building + AI coaching = a unique point of view on how AI changes careers.
ACT I
WE'VE BEEN HERE BEFORE
Bennis & Nanus (1985) / U.S. Army War College (1987) ↗
Volatility
Rate of change accelerates beyond prediction.
Claude Cowork launch wiped $1T from software stocks in one week. Two AI crashes in 13 months. TCS, Infosys hit.
Uncertainty
Outcomes become impossible to forecast.
Hinton says 5 years to AGI. LeCun says decades. OpenAI bets $74B it's sooner. The creators can't agree.
Complexity
Integrated systems create cascading effects.
246K tech layoffs in 2025. S&P 500 up 18%. Same companies. Same year. India IT: 25K fired + 35K AI hires at the same firms.
Ambiguity
Cause and effect become unclear.
Individual workers report 40% productivity gains (McKinsey). Economy-wide data shows near zero (Goldman, NBER). Same tool, two scales, opposite answers. Is AI creating value or just reshuffling it?
"The pace of change has never been this fast, yet it will never be this slow again."
Justin Trudeau, Davos 2018 · S&P 500 avg company lifespan: 67 years → 15 years (Innosight / McKinsey)
Satya Nadella
"Self-claiming some AGI milestone is just nonsensical benchmark hacking to me. The real benchmark is the world growing at 10%."
Satya Nadella CEO, Microsoft · Dwarkesh Podcast, Feb 2025 ↗
Y = A · Kα · L(1-α)
Cobb-Douglas Production Function ↗
Y Output
Total GDP. Everything society produces.
A Total Factor Productivity
Fundamentally: technology & efficiency. The multiplier that drives growth. AI changes this.
K Capital
Machines, money, infrastructure. Concentrating fast.
L Labor
Human work. AI is replacing this.
For 250 years, technology (A) amplified labor (L). AI is the first technology that substitutes for it.
MECHANIZATION1771 - 1829 STEAM & RAILWAYS1829 - 1873 STEEL & ELECTRICITY1875 - 1929 OIL & MASS PRODUCTION1908 - 1974 INFO & TELECOM1971 - 2020 AI2022+ 17701820187019131950197320002020$1K$2K$5K$10K$20KYEARGDP PER CAPITA (LOG) Canal Mania Railway Mania Panic of 1893 Wall St. Crash Dotcom Crash DeepSeek Shock? Maddison Project GDP per capita (2011 USD) · Carlota Perez Framework
EVERY WAVE CAUSED PANIC
Each time, new jobs emerged that nobody predicted: software engineers, data analysts, UX designers.
The Lump of Labor Fallacy
The mistaken belief that there is a fixed amount of work in the economy. Economists have argued for over a century that technology always creates more jobs than it destroys. Is this time different?
Wikipedia: Lump of Labour Fallacy ↗
Technology makes it easy for everyone to do a particular job. Once it is democratized, we all abstract a level higher to do a different kind of work.
Yesterday
Typing
Very few people were professional typists. Today, everyone types on their phone. You don't need someone else to type for you anymore.
Photography
Professionals operated expensive equipment. Today, everyone is a photographer. Pros shifted to art direction, storytelling, brand.
Today
Engineering
AI is democratizing software. Anyone can build an app or a product. The cost of building is approaching zero.
The New Expectation
Delivery is no longer the bottleneck. You need strategic vision, product thinking, understanding of competitive landscape, and a vision for where the product should go.
"The skill isn't building anymore. The skill is knowing what to build and why."
Cognitive
Replaces thinking itself. Analysis, judgment, creativity, strategy.
Information
Replaced clerical work. Filing, calculation, communication.
Electrical
Replaced repetitive operations. Assembly lines, powered grids.
Physical
Replaced manual labor. Machines did the heavy lifting.
Each revolution let humans climb to higher-order work. AI targets the highest order.
U.S. Employment Share
Data spans 1800 – 2024 · Bureau of Labor Statistics
Agriculture
70%(1800) 1.6%(2024)
Manufacturing
33%(1950) 8%(2024)
Services
15%(1800) 84%(2024)
???
What's next?
ACT II
IS THIS TIME DIFFERENT?
If you don't need
intellectual labor
anymore...
What's left?
Every technology revolution has created new jobs to replace the ones it destroyed. Economists call the belief that there's only so much work to go around the "lump of labor fallacy." But this time, the debate is genuinely open.
Sam Altman
Sam Altman
CEO, OpenAI
"AI will create new kinds of jobs we can't even imagine. The economy will grow. Everything gets cheaper."
Davos, Jan 2025 ↗
Sam Altman
Sam Altman
Same person, different day
"Jobs are definitely going to go away, full stop." AI will replace "95% of what marketers, strategists, creative professionals use agencies for."
Reddit AMA, Sep 2024 ↗
Daron Acemoglu
Daron Acemoglu
MIT · Nobel Prize 2024
"If we go down this path of destroying jobs, U.S. democracy is not going to survive." Warns of "so-so automation".
NYT / Nobel Prize 2024 ↗
David Autor
David Autor
MIT Economics
"AI can democratize expertise: nurses doing what doctors do, paralegals doing what lawyers do. This could rebuild the middle class."
NBER Working Paper, 2024 ↗
The honest answer: nobody knows. Even the same person contradicts themselves. That's what makes this moment so important.
Frontier Models · March 2026
Multiple models now exceed estimated human expert ceilings on science benchmarks (GPQA human expert ~80%, models at 91-94%)
Claude Opus 4.6
Anthropic · Feb 2026
80.8%
SWE-bench
91.3%
GPQA Diamond
GPT-5.2
OpenAI · Dec 2025
100%
AIME 2025 (math)
93.2%
GPQA Diamond
Grok 4.20
xAI · Feb 2026
1535
Arena ELO (provisional)
4.2%
Hallucination rate
Software Engineering Benchmarks · SWE-bench Verified
Opus 4.6
80.8%
GPT-5.2
80.0%
Gemini 3
76.2%
GPT-5
74.9%
o1 (Sep '24)
48.9%
From 49% to 81% in 18 months. Software is ground zero for AI capability.
REAL-WORLD SIGNAL
41% of all code written in 2025 was AI-generated. 46% of GitHub Copilot output gets accepted. Google + Microsoft: ~25% AI-assisted code each. 95% of engineers now use AI tools weekly.
Sources: GitHub, Google, Pragmatic Engineer Survey 2026
Benchmark Saturation · Are We Running Out of Tests?
GSM8K 99% (math)
MMLU 93% (knowledge)
HumanEval 98% (code)
GPQA 92% (PhD science)
AIME 100% (competition math)
All saturated. So researchers built harder tests. Even those are falling fast.
Humanity's Last Exam ↗
2,500 Qs from 1,000 experts across 50 countries · Nature 2025
One year ago: ~3%
Now: 53%
Opus 4.6
53.1%
Gemini 3 DT
48.4%
GPT-5.3
39.9%
FrontierMath Tier 4 ↗
Research-level math by Epoch AI · "Requires an entire math dept"
Only 17/48 open
problems solved
GPT-5.2 Pro
31%
Gemini 3 Pro
26.6%
GPT-5.2
14.6%
Dario Amodei
"We could have models that are a country of geniuses in the data center in one to two years."
Dario Amodei CEO, Anthropic · Dwarkesh Podcast, Feb 2026 ↗
WHAT THESE MEASURE
SWE-bench ↗ Real GitHub bugs. Can AI fix a codebase?
GPQA Diamond ↗ PhD-level science Qs. Experts get 30% wrong.
AIME ↗ Elite math competition. Top 5% qualify.
Arena ELO ↗ Crowd-ranked head-to-head. Like chess ratings.
LiveCodeBench ↗ Fresh coding problems, no memorization.
Humanity's Last Exam ↗ 2,500 Qs from 1,000 experts. Best: 53%.
FrontierMath T4 ↗ Research math. "Requires a math dept." Best: 31%.
The jump is exponential
Bar Exam: GPT-3.5 → 10th %ile. GPT-4 → 90th. One generation.
GPQA expert ceiling: ~80%. Models now: 91-94%.
BENCHMARK SATURATION
MMLU: 4 yrs · GPQA: 2 yrs · MMLU-Pro: 1.5 yrs
Each new benchmark falls faster.
PAGE 1 / 3
Click → to scroll
L1  Conversational
ChatGPT, Claude · Nov 2022
L2  Reasoning
o1, o3, DeepSeek R1 · Sep 2024
L3  Agents
WE ARE HERE · 2025
L4  Innovators
Novel discoveries · ~2027-28?
L5  Organizations
Full autonomy · ~2030s?
L1 → L2: 2 years · L2 → L3: 1 year · Accelerating
Kurzweil's Law of Accelerating Returns Each paradigm shift happens faster than the last
Time (log scale: billions of years → years → months) Rate of Paradigm Shifts Life 3.8B yrs Multicellular 1B yrs Homo Sapiens 300K yrs Agriculture 10,000 BC Writing 3,500 BC Printing Press 1440 Industrial Rev. 1760 Electricity 1880 Computer 1945 Internet 1995 AI 2023 2045? WE ARE HERE
Compute per dollar: 20 quadrillion-fold increase over 120 years
2029: Turing test · 2045: Singularity (Kurzweil, 2005)
L1 Conversational
L2 Reasoning
L3 Agents
L4 Innovators
L5 Organizations
WE ARE HERE · 2025
What is an agent?
A system that repeatedly pursues a goal by:
i) Planning and Reasoning
Breaking down complex goals into steps, evaluating options, making decisions
ii) Acting on the Environment using Tools
Writing code, searching the web, calling APIs, sending messages, editing files
iii) Context and Memory
Retaining knowledge across interactions, learning from past outcomes, building on prior work
AI agents are already here
Coding Agent
Reads specs, writes code, runs tests, fixes bugs, deploys
Claude Code Cursor Copilot Devin
Research Agent
Searches, reads papers, synthesizes findings, generates reports
Perplexity Deep Research Elicit
Customer Service Agent
Handles queries, resolves issues, escalates when needed, 24/7
Sierra AI Intercom Fin Zendesk AI
Data Analysis Agent
Queries databases, finds patterns, creates visualizations
Julius AI Databricks Copilot
Creative Agent
Generates copy, designs, produces content, iterates on feedback
Jasper Canva AI Midjourney
Information Work Agent
Drafts documents, manages email, schedules meetings, summarizes threads
Claude Cowork Microsoft Copilot
ACT III
GROUND ZERO: SOFTWARE
AI adoption is not uniform across industries. Software engineering leads by a wide margin, but the pattern is diffusing outward.
AI Task Automation Potential by Industry
Software & IT
85%
Legal
62%
Finance
58%
Marketing
52%
Education
42%
Healthcare
35%
Manufacturing
25%
Sources: Anthropic Economic Index, McKinsey Global Survey 2025, Stanford AI Index 2025
Why does software lead?
Work is 100% digital. The entire signal-to-closure loop exists as data. Decades of training data (code, docs, git history) available.
What happens in software today will happen everywhere in 3-5 years.
"Understanding what's happening in software right now is understanding everyone's future."
FORTUNE Feb 2026
Block lays off 4,000 of 10,000+ employees, blames AI
Dorsey: "Most companies will make similar cuts in the next year."
Read article ↗
CNBC May 2025
Klarna CEO: AI helped us shrink workforce by 40%, from 5,000 to 3,000
Projects under 2,000 by 2030. AI handles 50% of all customer service.
Read article ↗
CNBC Sep 2025
Salesforce CEO confirms 4,000 layoffs: "I need less heads with AI"
AI agents now handle 50% of all customer interactions. Benioff celebrated the milestone.
Read article ↗
CLOSER TO HOME
CNBC Aug 2025
TCS announces 12,000 job cuts, its biggest ever
Primarily mid and senior management, as AI coding agents do similar work at marginal cost.
Read article ↗
BUSINESS TODAY Feb 2026
$100 billion wiped from Indian IT market cap in one year
75 firms. Nifty IT stocks fell 11-28% in a single month. FPIs dumped Rs 10,965 crore in 15 days.
Read article ↗
STORYBOARD18 Jul 2025
TCS, Infosys, Wipro, HCL, Tech Mahindra shed 57,891 employees in two years
Revenue per employee rising across the sector. Fewer people, more output. The outsourcing model is being rewritten.
Read article ↗
These aren't predictions. These are reports from the field.
Andrej Karpathy
@karpathy
"I've never felt this much behind as a programmer. The profession is being dramatically refactored... some powerful alien tool was handed around... the resulting magnitude 9 earthquake is rocking the profession."
Dec 27, 2025 · View on X ↗
Boris Cherny
Creator of Claude Code, Anthropic
"In the last 30 days, I landed 259 PRs, 497 commits, 40k lines added. Every single line was written by Claude Code. I shipped 22 PRs yesterday and 27 the day before, each one 100% written by Claude."
Jan 29, 2026 · Fortune ↗
Andrej Karpathy
@karpathy
"There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists."
Feb 2, 2025 · 4.5M views · Entered Merriam-Webster · View on X ↗
Dario Amodei
CEO, Anthropic
"AI is writing 90% of the code." (March 2025) Six months later: "Within Anthropic, that prediction is absolutely true."
March 2025 · Yahoo Finance ↗
Google: 30%+ of code AI-generated (Pichai, Q1 2025)
95% of engineers use AI tools weekly (Pragmatic Engineer, 2026)
"The cost to build software has come down to zero."
Amjad Masad, CEO of Replit
25%
of YC W25 startups have codebases that are 95% AI-generated
TechCrunch, Mar 2025 ↗
30%+
of all new code at Google is now AI-generated
Pichai, Q1 2025 Earnings ↗
~100%
of code at Anthropic is now written by AI
Anthropic CPO, 2025 ↗
46%
of code written by GitHub Copilot for its 20M users. 90% of Fortune 100 adopted.
GitHub, Jul 2025 ↗
$2B
Cursor ARR. Doubled in 3 months. Valued at $29B. Fastest-growing dev tool ever.
TechBuzz, Feb 2026 ↗
$456K
ARR app built by a non-developer using AI tools alone. 10K+ users in 3 months.
State of Vibecoding, Feb 2026 ↗
Single engineers now spin up 20 AI agents and ship what used to take a team of 40.
Fewer engineers needed
Single engineers with AI agents replace entire teams. Companies need 1/3 the headcount for the same output.
Roles are merging
Designers write code. PMs create prototypes. Engineers handle product strategy. Specialization is losing value.
The "Builder" emerges
One person who can ideate, design, code, ship, and market. Breadth matters more than depth.
This entire deck was built through vibe coding. I wrote zero lines of code. I have no training in writing code.
Boris Cherny
Creator of Claude Code, Anthropic
LENNY'S PODCAST
"By the end of the year, the title software engineer is going to start to go away and be replaced by builder."
Feb 2026 · Fortune ↗
N
Naval Ravikant
@naval
"Vibe coding is the new product management. Training and tuning models is the new coding."
Feb 2026 · View on X ↗
G
Garry Tan
CEO, Y Combinator
"In the age of prompting and vibe coding, everyone's a PM now."
May 2025 · Vanta ↗
J
Jackie Bavaro
Author, "Cracking the PM Interview"
"Vibe coding is a dream come true for PMs: all the skills we've built working with engineers are exactly what you need to work with an AI teammate."
A product manager controls risk while enabling innovation. The biggest risk was always cost of production: 10 ideas, limited engineers, 3-5 month estimates, ruthless prioritization.
That constraint just disappeared.
OLD PROCESS
Idea PRD Debate & Refine Prioritize 3-5 months Ship
M
Madhu Gurumurthy
Head of Product, Google Gemini
"At Google, we are moving from a writing-first culture to a building-first one. Writing was a proxy for clear thinking, optimized for scarce eng resources and long dev cycles. Now, when time to vibe-code prototype ≈ time to write PRD, PMs can SHOW not tell."
View on X ↗
WHAT FALLS AWAY
Resource allocation juggling. Stakeholder management overhead. Influence without authority politics. These were consequences of scarcity.
WHAT REMAINS
Deep customer understanding. Long-term strategic vision. Knowing which problems are worth solving. Building moats in a zero-cost world.
Delivery doesn't differentiate anymore. Vision does.
: The Human Edge
YESTERDAY
Engineering code, architecture Design UI, UX, visual Marketing growth, brand Finance budgets, ROI Growth scale, metrics PM coordinates Talked to everyone. Owned nothing.
TODAY
codes designs markets analyzes scales Builder ideate · design · code · ship · market Does everything. Owns everything.
COMPANIES NOW REQUIRING PMs TO VIBE CODE
Google
Vibe coding interview round for AI PMs
Meta
PMs vibe code prototypes for Zuckerberg
Microsoft
AI fluency expected across all PM levels
Stripe · Netflix · Figma · Perplexity · Vercel
Adding AI prototyping to PM hiring
If you can vibe code it, it's feasible. The prototype becomes the starting point.
Carnegie Mellon now teaches vibe coding to PM students · Collins Dictionary 2025 Word of the Year: vibe coding
Agency + Taste
Direction, judgment, and aesthetic sense. The why, not the how.
Cognitive
Replaces thinking itself. Analysis, judgment, creativity, strategy.
Information
Replaced clerical work
Electrical
Replaced repetitive operations
Physical
Replaced manual labor
Each revolution displaced a human capability. AI displaces cognition. What remains above it is yours.
Garry Tan
Garry Tan @garrytan
CEO, Y Combinator
"Intelligence is on tap and humans must provide the agency and taste."
Paul Graham
Paul Graham @paulg
Co-founder, Y Combinator
"When anyone can make anything, the big differentiator is what you choose to make."
Sam Altman
Sam Altman @sama
CEO, OpenAI
"Agency, willfulness, and determination will likely be extremely valuable."
The cost of producing the asset is near zero. So the bottleneck has shifted: from building to distributing.
Blue Ocean
Uncontested market space where demand is created, not fought over. No competition yet.
Red Ocean
Existing market with defined boundaries, accepted rules, and companies fighting over shrinking demand.
AI turns every blue ocean red before you can finish swimming. When building costs nothing, barriers to entry collapse, and every market floods with competitors instantly.
"The new job for everyone has become marketing."
We started by saying software is ground zero. AI has completely disrupted ways of working and the means of production inside the software industry.
But make no mistake: any role where 100% of work is digital is already in the disruption zone. Financial analysts, content writers, designers, legal researchers. If your entire workflow exists as data, AI can already do your job.
For industries that involve the physical world, two barriers remain unsolved:
MR
Mike Rowe
Host, Dirty Jobs · mikeroweWORKS
"We've been telling kids for 15 years to learn to code."

"Well, AI is coming for the coders."

"It's not coming for the welders, the plumbers, the steamfitters, the pipefitters, the HVAC, or the electricians."

"In Aspen, I sat and listened to Larry Fink say we need 500,000 electricians in the next couple of years."
PA Energy & Innovation Summit, Jul 2025 · View on X ↗
Context
Organizations have decades of institutional knowledge, workflows, and decision-making patterns locked in people's heads. Collecting this context for AI is the #1 challenge.
Physical World
AI needs to sense and act in the real world. That's where robotics comes in.
Tesla Optimus
Factory floors
Unitree
Agile humanoids
Boston Dynamics
Advanced mobility
Solving these barriers is the biggest market opportunity of the next decade.
The same pattern from software is repeating everywhere: costs collapse, roles merge, fewer people produce more.
LEGAL
Harvey AI
$8B valuation. 50 of top 100 law firms. Lawyers save 4 hrs/week, worth $100K in billable time per lawyer annually.
"2026 is the year firms measure cost savings." · Harvey Blog ↗
MEDIA
Pocket FM
90% of content is AI-generated end-to-end. $350M ARR. Content costs down 2-3x. Pilot creation time reduced 90%.
CEO Rohan Nayak, TechSparks 2025 · YourStory ↗
MARKETING
Content Collapse
Writer roles -28%. Graphic artists -33%. Photographers -28%. 80% of marketing leaders now use AI for content.
Bloomberry, 180M job postings analysis · Source ↗
HEALTHCARE
Insilico Medicine
First AI-discovered drug to Phase II. Discovery in 30 months vs. 4-6 years. AI radiology: 94% accuracy vs. 65% human.
Nature Medicine, Jun 2025 · Insilico ↗
FINANCE
Goldman + Devin
Deployed Devin as "first AI employee." 54% of banking jobs at high automation potential. The $180K engineer is being replaced.
Citi Report · CNBC ↗
EDUCATION
Chegg: -99%
Stock crashed 99% post-ChatGPT. 67% of workforce cut. Duolingo replaced translators with AI, calls them "content curators" now.
Gizmodo ↗ · Fortune ↗
Same pattern. Every industry. Costs collapse. Roles merge. Fewer people, more output.
ACT IV
WHO CAPTURES THE VALUE?
306
S&P 500 companies cited AI on
earnings calls (Q3 2025)
3.6x the 10-year average
+8.2pp
Stock outperformance of companies
that mention AI vs. those that don't
FactSet, YTD Q3 2025
44%
of S&P 500 now list AI expertise
in director qualifications
Up from 26% one year ago
Jensen Huang
Jensen Huang
CEO, NVIDIA
"You're not going to lose your job to an AI. You're going to lose your job to someone who uses AI."
All-In Podcast, Jul 2025
Marc Benioff
Marc Benioff
CEO, Salesforce
"We are the last generation of CEOs to manage only humans. Every CEO in the future will manage both humans and agents."
Davos, Jan 2025
Jamie Dimon
Jamie Dimon
CEO, JPMorgan Chase
"There will be no job, no process, no function that won't be affected by AI. It's about getting people to understand the power of it."
JPM Investor Day, Feb 2026
The pressure flows downward: investors, boards, CEOs. The opportunity flows to whoever figures out how to diffuse AI across the organization.
: For You
FOR ORGANIZATIONS
Bottom-Up Learning Wins
Top-down AI training fails because every function has unique organizational context that no external consultant can replicate.
The most effective approach: give employees dedicated time, AI compute access, and aligned incentives to build agentic workflows that improve their own work.
78%
of AI users bring own tools without approval
95%
of gen AI pilots fail to scale beyond experiments
Manage Human + AI Teams
Managers must be empowered to build systems where they orchestrate both AI agents and human employees.
Org charts will evolve from hierarchies to agentic networks, where work is exchanged through tasks and outcomes, not reporting lines.
HBR: "Agent Manager"
A permanent new role. Within 12-18 months, a standard title in AI-first enterprises.
Embed First, Optimize Later
We have decades of process design that tells us why roles exist: who builds, who reviews, who approves. These touchpoints were created with intent.
Step 1: Embed AI into each existing touchpoint. Leverage proven process wisdom.
Step 2: Over time, identify opportunities to make workflows truly AI-native.
"AI is 20% algorithms, 80% organizational rewiring."
Lead First
Leaders must model AI use themselves. Vacate time, experiment openly, set the cultural tone from the top.
Allow Open Exploration
Early tasks may seem trivial. Comfort with the tool precedes solving high-value problems.
Grow Internal AI Explorers
Internal practitioners with org context outperform external consultants. They spot deployment opportunities and diffuse knowledge organically.
High-Friction First
Start where AI removes daily pain points. Then close the loop on end-to-end workflows for real productivity gains.
Build Knowledge Networks
Create mechanisms for teams to share AI context, patterns, and wins across the organization.
Build Evaluation Muscle
Enable employees to create agent evaluation mechanisms. Measure performance and quality of agentic workflows systematically.
Default to Agentic-First
Cultivate the instinct to ask "Can an agent do this?" before "Who should I assign this to?"
Redefine the Manager Role
Agent managers don't just delegate. They orchestrate workflows, set guardrails, and quality-check agent outputs.
Rethink Team Boundaries
Agents work across functions. Org structures must evolve from silos to fluid, task-based networks.
Unblock Individuals First
Ensure every employee can augment their own work before optimizing cross-team workflows.
Then Solve Cross-Team Friction
Build new playbooks for interconnected human + AI teams to move faster together.
Design AI-Native Workflows
Don't just layer AI onto old processes. Rethink workflows where humans and agents each play to their strengths.
Iterate in Production
Perfection delays adoption. Ship embedded AI early, gather feedback, refine. Speed of learning beats quality of planning.
88% of organizations use AI. Only a third have scaled it. The gap is not technology. It's management.
THE SKILLS THAT MATTER
AI Agent Literacy
Like organizational behavior gave us frameworks for managing humans, you need deep understanding of how AI agents work, what they're good at, and where they fail.
Diagnose failures. Design guardrails. Know when to trust, when to verify.
Systems Thinking
You're no longer doing the work. You're managing interconnected systems that span functions and disciplines. Seeing how components interact matters more than mastering any one.
Vision Over Execution
Execution is commoditized. Any agent can execute. The value lies in how superior and well-thought-out your vision is compared to someone else's.
KNOW THE LIMITS
Karpathy's 6 Cognitive Deficits of LLMs
"LLMs are fallible people-spirits. Their output must be verified, not trusted outright."
1
Hallucinations — They make things up. "Hallucination is all LLMs do. They are dream machines."
2
Jagged Intelligence — Solve complex math, then claim 9.11 > 9.9. Capabilities are deeply uneven.
3
Poor Self-Knowledge — Cannot assess what they know vs. don't know. Confident on topics where they should express uncertainty.
4
Anterograde Amnesia — "A coworker who forgets everything after each conversation." No long-term memory consolidation.
5
Gullibility — Susceptible to prompt injection. Can be tricked into leaking data or giving harmful responses.
6
Spinning — Vague prompts cause repetitive correction loops. Gets stuck rather than making progress.
Andrej Karpathy, YC AI Startup School ↗
"AI will not deliver value because firms spend money on tools. It will deliver value when leaders develop the competencies to transform organizations."
Harvard Business Review
Not the model. Three things matter:
Compute
Raw processing power at scale
Context
Proprietary data and domain knowledge
Distribution
Reaching users where they already are
ACT V
YOUR MOVE
01 PROMPTING
Instruct AI with precision. Clarity is the new management skill.
02 EVALUATIONS
Measure AI output quality. If you can't evaluate, you can't improve.
Andrej Karpathy
Andrej Karpathy @karpathy
Ex-Tesla AI, Ex-OpenAI
"The hottest new programming language is English."
Sam Altman
Sam Altman @sama
CEO, OpenAI
"Figuring out what questions to ask will be more important than figuring out the answer."
Garry Tan
Garry Tan @garrytan
CEO, Y Combinator
"Being able to do evaluations of what models and what prompts are good... That's actually turning out to be the moat for many startups."
poly·math /ˈpɒlɪmaθ/ — person of wide knowledge across multiple fields.
Leonardo da Vinci: painter, engineer, anatomist, architect, inventor. The original one-person team.
→ AI is making this achievable for everyone.
WHAT WE'RE SEEING
Roles Are Fusing
Specialists built the old world. AI collapses boundaries between disciplines. A single person now has the toolkit of an entire team: design, code, market, analyze.
Abstraction Upward
Entry-level execution is being filled by agent workforces. Humans move upward to strategy, judgment, and orchestration. The floor rises.
The Gap Narrows
As agents handle more of the value chain, the distance between capital owners and the working individual (labor) shrinks dramatically. Everyone becomes closer and closer to becoming a capital owner itself.
THE OPPORTUNITY
The One-Person Billion-Dollar Company
A single founder with AI agents can ideate, build, ship, market, and scale. The generalist who understands the whole system wins.
Sam Altman
Sam Altman @sama
CEO, OpenAI
"...there's this betting pool for the first year that there is a one-person billion-dollar company."
Alexis Ohanian
Alexis Ohanian @alexisohanian
Co-founder, Reddit
"...a single-person company had no chance of reaching a billion-dollar valuation. But now thanks to all this AI, it is."
WHERE WE'RE HEADED
L5 OrganizationsAI runs end-to-end
L4 Innovators~2027-28
L3 Agents◀ WE ARE HERE (2025)
L2 Reasoners2024
L1 Chatbots2023
"I skate to where the puck is going to be, not where it has been."
Wayne Gretzky
2023 2030+
We can't optimize for where we are today. We have to prepare for where AI will take us.
AND THAT BRINGS US TO THE ABUNDANCE ERA
When AI reaches L4-L5, it doesn't just execute. It innovates. Problems unsolved for decades become tractable. Technology compounds so fast that our ability to address humanity's biggest challenges explodes.
Sam Altman
Sam Altman @sama
CEO, OpenAI
"In the 2030s, intelligence and energy are going to become wildly abundant. The ability for one person to get much more done will be a striking change."
"We won't experience 100 years of progress in the 21st century. It will be more like 20,000 years of progress."
Ray Kurzweil, Law of Accelerating Returns
Each of you will have the opportunity to find a unique value proposition, build your own AI organization, and solve problems that matter. The problem space is no longer closed. It's wide open.
AI Budget
Allocate personal time and money to AI tools. Treat it as an investment, not an expense.
Build AI Fluency
Use AI daily. Push its limits. Break it. Learn what it can and cannot do.
Think in Workflows
Decompose your job into steps. Identify which steps an agent can own entirely.
Become a Generalist
The age of deep specialization is ending. Learn to orchestrate across domains.
"So let me ask again... do you feel anxious?"
Every previous revolution replaced muscle. This one replaces mind. We have never been here before. But if you have taste, instinct, and the willingness to explore, this is the greatest era to be alive.
Explorer
Find problems worth solving.
Chase curiosity relentlessly.
Builder
Ship what you imagine.
The tools are free. Start today.
Innovator
Solve what couldn't be solved.
This is your century to shape.
The moment is yours, the only variable is you.
/ 35