Dr. Harish Kotadia Ph.D.

Why I Wrote Agentic AI Book Earned Autonomy
Linkedin
August 01, 2026

I wrote Earned Autonomy because enterprise agentic AI has no shared instrument for answering the only question that matters at the point of deployment: how much autonomy has this agent actually earned? The book supplies one. It is a governance methodology built around a 0–1000 score across five pillars and twenty criteria that turns an agent's authority from a claim into a measurement a risk committee can read, challenge and revoke.

Agentic AI Just Undid 30 Years of IT Discipline
Linkedin
July 26, 2026

Originally published by Dr. Harish Kotadia, Ph.D. on AgenticAIArch.com: https://agenticaiarch.com/agentic-ai-just-undid-30-years-of-it-discipline/

Vibe coding is having its moment. But last July during a Vibe coding event, an AI agent deleted a live production database during a declared code freeze, then told its user that rollback was impossible. It wasn’t.

That is not a model failure. It is a missing dev/prod boundary, a missing approval gate and a missing rollback path — three controls IT shipped way back in the 1990s.

Here are ten controls Agentic AI quietly undid, that enterprise IT spent thirty years getting right.

What Is Agentic AI Technology? A Practitioner’s Field Map of the Enterprise Stack
Import from wordpress feed
July 19, 2026

“Agentic AI technology is the software stack that lets an enterprise hand a system an outcome to achieve rather than a script to execute”  © Dr. Harish Kotadia, 2026 — the model, memory, tools, identity, guardrails, and observability that turn a goal into governed, auditable action.

Token Troubles: Five Enterprises That Learned the Hard Way
Linkedln
July 17, 2026

That is the paradox defining agentic AI economics in 2026, and it is not a pricing problem. It is an architecture problem. A chatbot answers a question with one inference call. An agent decomposes a task, selects tools, calls sub-agents, validates outputs, retries on failure — and every one of those steps burns tokens. A leading industry analyst firm's March 2026 analysis found agentic workflows consume 5 to 30 times more tokens per task than a standard chatbot query. Some research puts the multiplier far higher.

Why Agentic AI Budgets Are Surging in 2026 — and the Fix
Import from wordpress feed
July 13, 2026

Agentic AI budgets are breaking in 2026, and not because the models are bad. They are breaking because enterprises are deploying agents without mapping token spend to projects and outcomes. Agentic AI cost attribution is the practice of tagging every token an agent consumes to a named project, workf

Why I Wrote This Book on Agentic AI
Import from wordpress feed
July 12, 2026

Over the past eight months, I have had the opportunity to speak with senior technology professionals at Fortune 500 companies while evangelizing agentic AI solutions. What I realized surprised me: many of them were not doing agentic AI right. Not because of lack of talent or budget, but because they

Two Agentic AI Deployments, One Contained, One Catastrophic
Import from wordpress feed
July 07, 2026

The Difference Was Operational Risk Management Summary: Of the two 2025 agentic AI coding deployments compared here, one sits at Stage 3, Governed, on my Agentic AI Roadmap, and one operated at Stage 2, Piloted, while being marketed as Stage 5, Autonomous. The gap between them was not model capabili

11 OpenAI Agentic AI Case Studies Analyzed and Mapped
Import from wordpress feed
July 06, 2026

Quick answer Of the 11 OpenAI case studies analyzed here: 1 sits at Prompted, 3 at Piloted, 3 at Governed, 3 at Assured, and 1 at Autonomous. This is the most evenly distributed maturity spread across OpenAI, AWS, and Claude vendor case studies analyzed in this series. Companies that reach Assured o

Agentic AI Autonomy Trap: Healthcare vs Fintech Case Study
Import from wordpress feed
July 05, 2026

Two Agentic AI Case Studies – One Success, One Reversal And What Separates Them I have spent this year documenting agentic AI wins. This edition is different. I want to put a genuine success story next to a genuine reversal, side by side, because the gap between them is not about the technolog

Agentic AI’s Autonomy Trap – A Case Study
Import from wordpress feed
July 05, 2026

I have spent this year writing about agentic AI wins across banking, healthcare, and manufacturing. A European fintech company is the case I keep coming back to for a different reason. It is the clearest public record of a company that pushed straight past Governed toward a claimed Autonomous, skipp

Agentic AI in Healthcare: The Prior Authorization Case Study
Import from wordpress feed
July 04, 2026

For as long as I have been building agentic AI systems, I have heard a version of the same objection when the conversation turns to healthcare. Protected health information cannot touch an AI agent, they say. The liability is too high, the compliance burden too strict, the margin for a confident wro

Agentic AI in Regulated Industries: Case Studies Proving Skeptics Wrong
Import from wordpress feed
July 03, 2026

The ABA Banking Journal recently reported on the broader shift now underway across financial services, from Anthropic’s ten ready-to-run agent templates for pitchbooks, KYC screening, and month-end close, to a major wealth management firm’s advisor note-taking agent, to another wealth ma

The Agentic AI Roadmap Stage Behind Tesla Robotaxi Headline
Import from wordpress feed
July 03, 2026

Tesla turned on unsupervised Robotaxi service in Miami. No safety monitor in the seat, live in a geofenced zone covering West Miami, Doral, and Coral Gables. Florida is now the first state outside Texas to get it, and Miami is the fourth city in the network after Austin, Dallas, and Houston. I read

From Siebel to Agentic AI: How Enterprise Data Finally Got Free
Linkedln
June 29, 2026

I have spent over twenty years installing software so that business people could look at their own customer data. I mean that literally. Some of my earliest projects involved putting a CRM client on individual desktops, one machine at a time, before a salesperson could see a single account they already owned. That should sound strange to you. It took me a long time to find it strange too.

11 AWS Agentic AI Case Studies Analyzed and Mapped
Linkedln
June 28, 2026

I spent this weekend going through the AWS Solutions Library and its customer success stories, hunting for the deployments that have actually shipped. I was specifically interested in AWS agentic AI case studies—not the polished demos, but the ones running in production with real numbers attached to them.

Agentic AI Assurance Is Now a Market, for REAL
Linkedln
June 26, 2026

Something new is showing up in enterprise AI budgets. Not agents. The audit of agents.

A market is forming around agent assurance. Independent third-party review. Runtime monitoring. Even guardian agents whose only job is to watch other agents and rein them in when they drift.

The Agentic AI Roadmap: 5 Levels From Prompted to Autonomous
Linkedln
June 26, 2026

Instructions in, results out. That was IT. Intent in, outcomes out. That is agentic AI. But an organization does not make that turn all at once. It matures into it. Forty years ago, software engineering needed a way to talk about that maturity and the Capability Maturity Model gave it five honest levels, from ad hoc to optimizing.

Who Was That? The Agent Identity Question Nobody Can Answer
Linkedln
June 25, 2026

For most enterprises, there is no good answer. In a 2026 survey the Cloud Security Alliance ran with Aembit, more than two-thirds of organizations could not clearly distinguish an action taken by an AI agent from one taken by a human. Sixty-eight percent. The agent does the work, and the log records a person. That is not a reporting gap. It is an accountability gap, and accountability is the whole game in a regulated workflow.

Agentic AI Reality Check: The 40% That Will Fail
Linkedln
June 24, 2026

The loudest number in agentic AI right now is a warning. Gartner says more than 40 percent of agentic AI projects will be canceled by the end of 2027. The reasons are not exotic. Escalating costs. Unclear business value. Risk controls that were never built. The same research calls out "agent washing," and estimates that only about 130 of the thousands of vendors claiming to sell agents actually do.

Agentic AI Case Study: Auto Loan Processing on Claude
Linkedln
June 18, 2026

I have spent close to two decades architecting enterprise systems for Fortune 100 clients. The auto loan platform I built around Anthropic Claude was different. The AI did not sit on the side as a feature. It sat in the middle as the operating model. Here is an honest account of how it came together, what it delivered, where it bit me and what I would change.

What is Agentic AI? Definition of Agentic AI
Linkedln
June 15, 2026

Agentic AI is the next enterprise workload abstraction: a governed, goal-driven software layer in which LLM-powered agents — equipped with memory, tools, and orchestration protocols — autonomously plan and execute multi-step business processes across your cloud, data, and application estate, the way containers once abstracted servers and pipelines once abstracted data movement.

Agentic AI: 50 Authoritative Definitions — A Curated Citation Compendium by Dr. Harish Kotadia, Ph.D.
Linkedln
June 09, 2026

What exactly is Agentic AI? Ask ten experts and you’ll get ten answers — and that’s not a bug, it’s the most important insight in this space right now. Multiple credible sources, from MIT Sloan Management Review to Gartner, explicitly acknowledge there is no single universally agreed-upon definition.

Instructions In, Results Out — That Was IT. Intent In, Outcomes Out — That’s Agentic AI
Linkedln
June 09, 2026

I know, because I went looking. Over the past few weeks, I systematically reviewed the top 50 definitions of Agentic AI published across peer-reviewed scholarly journals (IEEE Access, Springer’s Artificial Intelligence Review, MDPI Future Internet, F1000Research), the leading industry research firms (Gartner, McKinsey, BCG, Deloitte, MIT Sloan), major technology companies (IBM, Salesforce, Microsoft, Anthropic, Google), and top business publications (The Wall Street Journal, Associated Press, MIT Sloan Management Review).

The hidden cost of NOT going agentic is bigger than your AI budget.
Linkedln
June 08, 2026

Most enterprises are still running linear LLM chains and calling it "AI transformation."

The Architecture Beneath the Hype: A Strategic Briefing for CIOs & Technology Leaders
Linkedln
June 07, 2026

Three major technology shifts have reshaped enterprise IT over the past two decades. The current one — the move from stateless LLM calls to agentic AI architectures — is the most consequential. Most organizations have invested in the wrong layer. This article explains the structural difference, why it matters on AWS and Google Cloud and what CIOs should do about it now.

Agentic AI P&L — Edition 2
Linkedin
August 11, 2026

From $0.04 to $1.20 Now: The 30x Isn't for the Model, it's for Checking Model's Work.

Enterprise Agentic AI Case Studies — Edition 5
Linkedin
August 10, 2026

What Two Test Environments Taught Me About the Word "Sandbox"
This edition of Enterprise Agentic AI Case Studies grew out of a piece I wrote on my site about two frontier labs whose own test environments reached real companies in July 2026.

The Audit Trail Edition: A Record Your Agent Wrote Itself Is Not Evidence
Linkedin
August 10, 2026

In April 2026, a researcher found a flaw in a widely used AI app-building platform. A free account could read other users' source code, database credentials, and full AI chat histories. It took five ordinary API calls. No advanced hacking required. The access had been closed the year before. A backend change in February quietly reopened it. Nobody caught the reopening for weeks.

Agentic AI P&L — Edition 1
Linkedin
August 09, 2026

The price of inference fell by half and the bill still went up 125%
In the twelve months ending in December 2025, the cost per million tokens fell by roughly half while tokens consumed grew 4.5 times. Halve the price, multiply the volume by 4.5, and the spend line comes out at 2.25 times where it started. That is my arithmetic on two figures Bain published in June 2026, and it is the single most expensive misunderstanding in enterprise AI planning right now.

The Ai4 Edition: Every Session This Week Is Really Asking One Question
Linkedin
August 03, 2026

By Dr. Harish Kotadia, Ph.D.
Ai4 opens tomorrow, August 4, at The Venetian in Las Vegas. Three days, one main stage, dozens of tracks.

Last year's Ai4 asked what agents can do. This year's agenda asks something harder: how do you govern an agent that acts faster than a human can review it?

Enterprise Agentic AI Case Studies — Edition 4
Linkedln
July 31, 2026

Bottom line: I put all 507 publicly documented enterprise agentic AI deployments I have collected through the same twenty questions. Three quarters are still supervised pilots. None are autonomous. And in 257 of them — half — the thing holding the deployment back is not security, not permissions, not the model. It is that nobody would notice if the agent quietly got worse.

Ten Parameters Decide Whether an Agent Is Actually Working
Linkedln
July 30, 2026

Most agent programs report one number to the steering committee. Task success rate.

It is the flattering number and the one that says least about whether the deployment has earned the right to keep running.

An agent that succeeds sixty percent of the time on a single attempt is not a sixty-percent-reliable system that occasionally stumbles. It handles the same request correctly on Tuesday and incorrectly on Thursday and downstream that looks like an inexplicable one-off.

Enterprise Agentic AI Case Studies — Edition 3
Linkedin
July 28, 2026

Three Agents Now Change a Live Mobile Network
Bottom line: A tier-one European mobile operator has three AI agents running in production on its live mobile network. They spot crowd events, check whether the nearby cell sites can carry the load and reconfigure the network before customers notice. Handling an event went from hours to about a minute. What made this safe was not the model. It was sorting the work into two piles up front — what an agent may do alone and what needs a person to press the button. Intent in, outcomes out only holds if you can say which outcomes an agent was allowed to produce.

Enterprise Agentic AI Case Studies — Edition 2
Linkedin
July 27, 2026

The Control That Makes This Agent Defensible Is That It Cannot Ship Anything Itself
Bottom line: A systemically important retail bank runs an agentic system that reads more than 80 million fraud signals a day and writes new fraud-detection rules on its own — and every rule stops at a human analyst before it takes effect. Fraud losses fell more than 20% in H1 FY2026. What makes it defensible is that the agent's output is a proposal, never an action.

Edition 14 - The Delegated Authority Edition
Linkedln
July 27, 2026

Your agent can pass its power on. The only question is how much.

I define delegated authority as the limit on what an agent may hand to the next agent, workflow or process it calls. You set it by deciding what each handoff may carry.

Most teams state that limit in a design document. Very few enforce it inside the credential. Your Agent Can Hand On Authority You Never Gave It

In our first case study, someone opened a ticket on a public code project. The project ships a widely used open-source AI coding agent with over five million users.

Enterprise Agentic AI Case Studies — Edition 1
Linkedin
July 24, 2026

The 80% Gain Came From What the Agents Were Forbidden to Do, Not What They Were Allowed to Do
Bottom line: A Fortune Global 100 European insurer built a seven-agent claims system that cut settlement time by 80% and shipped in under 100 days — and the design decision that made it defensible was refusing to let any agent authorize a payment. In regulated environments, the constraint is the architecture. Intent in, outcomes out only counts if the outcome survives an examiner.

A Practitioner’s Field Map of the Enterprise Stack
Linkedin
July 19, 2026

By Dr. Harish Kotadia, Ph.D.

This is a condensed edition of my full field map. The complete version — with the full stack diagram, vendor examples, and governance checkpoints — is on my blog: What Is Agentic AI Technology? A Practitioner's Field Map of the Enterprise Stack

Watch Your Tokens: The One Number Every Agentic AI Practitioner Must Track
Linkedin
July 16, 2026

The Agentic AI Architect — 10th Newsletter Edition
If you work anywhere near agentic AI, token usage is the number that decides whether your project survives its own budget.

Cheaper Tokens, Bigger Bills: The Economics Problem in Enterprise Agentic AI
Linkedin
July 15, 2026

Token prices are collapsing, and enterprise AI bills are climbing anyway.

Fortune reported in June that token prices have dropped more than 90% since 2023 while LLM spending has doubled since late last year. Apollo chief economist calls it Jevons paradox: as tokens get cheaper, companies "run more AI agents, automate more workflows and generate more code."

Agentic AI Budgets: The Token Trap
Linkedin
July 14, 2026

Agentic AI budgets are breaking in 2026, and not because the models are bad. They are breaking because enterprises are deploying agents without mapping token spend to projects and outcomes.

Two AI Agent Identities: One Borrowed, One Issued
Linkedin
July 09, 2026

The difference was identity architecture. One AI agent ran on borrowed user tokens and leaked data across roughly 1,000 organizations for a month. Another got its own credentials, its own team-scoped instance, and its own human manager before it touched production.

Two Agentic AI Deployments – One Contained, One Catastrophic. The Difference Was Risk Management
Linkedin
July 07, 2026

I have spent this year documenting agentic AI wins and reversals. Last edition, I compared a healthcare success against a fintech failure and traced the gap to human-in-the-loop design. This edition looks at the other half of that governance story: what the agent is allowed to touch before a human ever sees its work. The same month in 2025 produced two coding agent deployments that answer that question in opposite ways.

One Success, One Reversal: What Separates Them
Linkedin
July 05, 2026

I have spent this year documenting agentic AI wins. This edition is different. I want to put a genuine success story next to a genuine reversal, side by side, because the gap between them is not about the technology. It is about one missing stage in the roadmap.

The Case Studies Edition: Seven Lessons from Production Agentic AI
Linkedin
July 02, 2026

Case studies get written after the fact, when the architecture already works and the hard decisions are easy to forget. I went back through five sources: two origination systems I worked on and a survey of eleven Claude deployments, a survey of eleven AWS deployments, and a fifty-case roadmap of the market as a whole. Seven lessons came out of it, and every one of them shows up more than once across the five.

What Agentic AI Really Is — and Why Doing It Right Is the Whole Game
Linkedin
July 01, 2026

Instructions in, results out — that was IT. Intent in, outcomes out — that is agentic AI.

That single line is the whole shift, and most enterprises are still on the wrong side of it. So in this edition I want to do three things: say what agentic AI actually is, explain why it is genuinely different from the AI you shipped last year, and make the case that how you build it — not which model you pick — is what decides whether it survives production. I will close with the five-stage roadmap I use to place any deployment on the map.

Agentic AI for the Enterprise: One Practitioner’s Field Notes
Linkedin
June 30, 2026

I have spent close to two decades building enterprise systems for Fortune 100 clients. I have watched three real shifts in that time, and I am convinced the one happening now is the largest of them. Over the past few weeks I have written it down in pieces. This is the through-line that connects all of them, written for the people who have to decide what to do about it: the CIO, the CTO, and the executives sitting in the room when the budget is approved.

A Practitioner's Definition of Agentic AI
Linkedin
June 30, 2026

Welcome to the first edition. I'll keep these short and useful: field notes from building agentic AI in production, mostly in regulated industries, for the people who have to answer for what these systems do. Let me start where every real conversation should — with what agentic AI actually is.

Earned Autonomy: A Governance Methodology and Scoring Framework for Agentic AI
Independently published (Amazon KDP)
July 30, 2026

Autonomy is not a setting you enable. It is a position your organization earns — and the evidence says almost nobody has earned it yet.

Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027, and it names the causes in a revealing order: escalating costs first, unclear business value second, inadequate risk controls third. Model capability is not on the list at all. Only 23 percent of organizations have scaled an agentic system anywhere in the enterprise. And two-thirds of the technology chiefs surveyed by the IBM Institute for Business Value report the same thing in different words: the authority they have handed to agents has outrun the accountability they can exercise.

That gap is what this book closes.

A stage-gated methodology, not a manifesto

Earned Autonomy sets out five stages — Prompted, Piloted, Governed, Assured, Autonomous — and assesses three tracks across them: governance, agent identity, and security and execution boundaries. Two meters run beneath every stage: what the work costs, and what it was worth. Each stage has a testable exit gate. Nothing advances on assertion.

An Index, and 507 deployments scored on it

The Earned Autonomy Index scores a deployment from 0 to 1000 across five pillars and twenty criteria, then applies one rule that changes the answer: your band is gated by your weakest pillar, not licensed by your highest total. Applied to 507 publicly documented enterprise deployments, three quarters sit in the Piloted band and not one reaches Autonomous. Two pillars — Outcome Assurance and Intent Specification — cap 470 of the 507. The field has built the controls that stop an agent doing damage. It has not built the controls that would tell anyone the agent has stopped doing its job.

Inside

The supervision gap: why authority outruns accountability, and what it costs
Agent identity: the control plane an examiner asks about first
The four-question control plane: what an agent may do, whose words it may obey, what it may hand on
The Token Trap: engineering agentic economics before the invoice arrives
Ten parameters for agent performance, and how each one gets gamed
The audit-survivability test: for any action any agent took, can you show who acted, on whose authority, and at what cost?
A self-assessment that gives you a score, a capping pillar, and a next move
Who it is for

CIOs, CTOs, enterprise architects, and risk and audit leaders who must decide how much autonomy to grant — and defend the decision afterwards.

I have spent more than two decades building enterprise systems for Fortune 100 clients, most recently the agentic AI platforms behind regulated auto loan and credit card origination. The methodology in this book is what that work taught me.

Companion volume to Intent In, Outcomes Out: A Practitioner's Field Guide to Agentic AI for the Enterprise.

Intent In, Outcomes Out: A Practitioner's Field Guide to Agentic AI for the Enterprise
Independently published (Amazon KDP)
July 10, 2026

A practitioner's field guide to building and governing agentic AI systems in the enterprise. Drawing on production experience in regulated auto loan and credit card origination, this book introduces the Five-Stage Agentic AI Roadmap and makes one argument: the model is a commodity — the loop, governed from day one, is the product.

Enterprise Agentic AI Case Studies
LinkedIn
July 24, 2026

I write Enterprise Agentic AI Case Studies, a LinkedIn newsletter that examines one publicly documented enterprise agentic AI deployment per edition. Each analysis covers the business problem, the agent architecture and stack, the governance and control model, the measured outcome, and what I would do differently. Sources are named and claims are traceable, drawn from my library of 507 verified enterprise agentic AI cases.

Agentic AI P&L
LinkedIn
July 10, 2026

Enterprise agentic AI, costed. Payback periods, hidden failure costs and the P&L math.