Agentic AI 101 starts with a punched card. In the early 1970s computing meant machine language, then BASIC and COBOL on an IBM 370, fed through a card reader. You wrote the program, ran the compiler, waited for the executable and hoped. Fifty years later a claims manager types what she needs into a short file and an agent builds it. Every step between those two moments moved the same boundary: who is allowed to tell a computer what to do.
That boundary is the whole of agentic AI 101. For twenty years only computer scientists could cross it. Then programmers did, then business users did through screens an IT department built for them. Now, with agentic AI, the business user crosses it directly. The thing she crosses it with is a file I have written about before, called intent.md.
The machine room, 1970 to 1985
IBM announced the System/370 on June 30, 1970, and for most of that decade the 370 was what “computer” meant in a large company. Programs went in on punched cards, 80 columns each, through a card reader. You ran the COBOL or FORTRAN compiler as a batch job, collected the printout, found a typo in column 12 and start again. A compile could take an afternoon.
The languages of that room were built for people who already understood the machine. BASIC came out of Dartmouth in 1964 to teach beginners. COBOL and FORTRAN carried the payroll and the physics. Pascal arrived in 1970 to teach structure. Then Dennis Ritchie’s C emerged at Bell Labs in 1972 and C++ followed in 1985. Each one moved the programmer a little further from the hardware. None of them moved the business user an inch. In the 1970s and 1980s, if you wanted a report, you filed a request with the data processing department and waited.
The programmer’s decade, 1985 to 1995
C and C++ turned programming into an industry. Software stopped being something a company’s own staff wrote for its own mainframe and became something you bought. Then Java shipped in 1995 with the promise of writing once and running anywhere, just as the internet opened. The applications that followed, from browsers to trading systems, were built in these languages by a new profession.
The boundary moved from computer scientist to programmer. That was a real shift. A programmer did not need to know how the 370 addressed memory. But the programmer still stood between the business and the machine. The business said what it wanted in a meeting. The programmer translated it into code. Months later, the business found out what had been understood.
The business application, 1993 to 2000
The internet revolution of the early 1990s gave those languages a job: business applications. Siebel Systems, founded in 1993, and Vantive, founded in 1990, sold packaged customer-relationship software that ran a sales floor or a call center. PeopleSoft paid about $433 million for Vantive in 1999, CFO reported. I spent years implementing systems like these. The business bought the package, and the IT department configured the screens.
So the business user finally touched the computer. But only through a screen an IT department had built, showing fields an analyst had chosen, producing reports a programmer had written. The e-business wave after 2000 put those screens on the web and handed them to customers too. The boundary moved again, from programmer to business user, but the user was still reading. Any agentic AI 101 has to start from that fact. She was not yet telling.
The agent, 2025 onward
Agentic AI moves the boundary the last step, and this is the lesson of agentic AI 101. The business user tells the computer what she needs, and the computer works out how. In my definition, “where traditional IT systems execute instructions, agentic AI executes intent.” The form the intent takes is not a meeting or a ticket. It is a file.
Anthropic’s playbook for an AI-native lifecycle puts that file first. It has five parts: the problem, the proposed outcome, the affected users and systems, the constraints and the open questions. In Anthropic’s worked example the author is a claims operations manager, not an engineer. I called it a regulated record in August, because it answers “who asked for this, why, and under which constraints.” A constraint in that file is “a business person’s limit, captured in a place an agent will read.” That sentence is the whole fifty years in one line.
What agentic AI 101 teaches about the boundary
Here is the pattern I see when I lay the eras side by side. Each wave removed one translator. The compiler removed the person who wrote machine code. Packaged software removed the in-house programmer for common work. The web screen removed the report request. Agentic AI removes the last translator, the person who turned a business ask into a specification. What it does not remove is the ask itself. Someone still has to say what they want, why, and within what limits. That person is now accountable for the file.
This is why I keep saying agentic AI 101 is a governance class before it is a technology class. When the business user writes the intent, the definition of done and the harness around the agent decide whether the outcome is acceptable. The code review moves from the diff to the intent. In my work in regulated loan origination, the underwriter’s limit goes into the file before the agent runs, and the agent cannot talk its way past it.
I have configured CRM screens in a decade when it belonged to IT. Now it belongs to the person with the problem, and that is the right place for it. Instructions in, results out was IT. Intent in, outcomes out is agentic AI, and the whole of this history was the walk from the first to the second.
Which translator still stands between your business users and the machine?
Subscribe to the Blog: I publish a new post every weekday on one agentic AI governance question. Enter your email and each post arrives in your inbox the moment it goes live.
My Books: Three books on agentic AI for regulated enterprises: Intent In, Outcomes Out, Earned Autonomy and The Governed Loop. Reading order, formats and buy links are on one page.
© Dr. Harish Kotadia, Ph.D., All Rights Reserved, 2026
Dr. Harish Kotadia, Ph.D., is an Enterprise AI Architect with 20+ years of IT consulting experience serving Fortune 100 clients, specializing in agentic AI systems built on Anthropic Claude, AWS Bedrock, and Google Vertex AI.
Disclaimer: This blog post is based on publicly available academic publications, vendor documentation, open standards, and news items from reputed media sources linked above. This post is intended for educational purposes, to help the enterprise agentic AI community build a shared vocabulary from public, authoritative sources.
Views and opinions expressed here are my own and do not represent those of any employer or client, past or present. The analysis presented is my independent interpretation of the published sources linked above and does not constitute legal, financial, or consulting advice of any kind.