Human judgement × AI capability.
AI will not replace people who work well with it. It will take over much of the work of people who don’t. We train both halves of the equation, and assess each one separately.
What people bring
- Strategy and vision
- Ethics and judgement
- Creativity and taste
- Empathy and trust
What AI brings
- Speed of execution
- Data analysis
- Scale
- Pattern recognition
The same four pillars, weighted by degree.
Shares of studio credits. Engineers lean on technology; MBAs lean on business. Both clear a floor in every pillar.
Build with technology
Python and JavaScript, data and SQL, AI engineering, agents, testing and verification
Build business
Customers, unit economics, selling, product and ventures
Understand the world
How one industry works, and responsible use of data and AI in it
Lead and communicate
English under pressure, specs, client conversations, ownership
The core: AI-native judgement. Specify what to build, verify that it works, decide what to do about it. It is not a fifth subject. Every artefact a student submits carries a verification log, and the log is graded in every pillar.
One studio a semester, built around real work.
Integrated three- to four-credit studios replace a stack of thin courses. A studio has a client, a brief, a deadline and a review, the way work does.
One integrated studio per semester, inside the regular timetable, supported by short self-paced modules.
University faculty set the academic frame. Practitioners bring live cases and review the work.
In-person studios with your cohort. Online is support, never the main event.
Course outcomes mapped to programme outcomes for the Board of Studies and NBA files.
A course asks
Did you attend, and can you recall it in an exam hall?
A studio asks
Did you ship something that works, can you prove it, and can you defend it without AI?
From chat to action.
Asking an AI a question is the easy part. The work employers now pay for is designing systems where an agent plans, acts through real tools, checks its own results and hands the hard cases to a person. Every student builds one, starting in Year 1.
Worked example: chase a clinic’s overdue invoices
What the agent does
Takes a goal a person sets: collect payments on invoices more than 30 days overdue, without upsetting long-standing patients.
What the student is graded on
Writing a goal precise enough to test, including what the agent must never do.
Thirteen competencies, four levels of evidence.
Each level is defined by work a reviewer can see, not by hours attended. B.Tech students climb all four over four years. MBA students start at Level 2 on the business rows and aim for Level 4.
Verification and debugging
Spots wrong AI output in a familiar domain
Writes tests and evaluations before trusting output
Finds the root cause in AI-written code without AI help
Sets the verification standard for a team
Employers see these as twelve first-round competencies on Roles and skills.
Every skill, tested twice.
With AI, students show they can direct it, check it and measure the result. Without it, they show they understand what they built.
| Skill | With AI: direct, verify, measure | Without AI: explain, debug, defend |
|---|---|---|
| Software | Ship a feature with an AI coding assistant, with tests that prove it works | Find and fix a bug in the same code, live, on paper or a bare editor |
| Data | Use AI to draft the analysis, then reconcile every figure to the source | Write the SQL by hand and explain why the numbers moved |
| Agents | Build an agent and an evaluation suite that catches its failures | Walk a reviewer through what the agent does when a tool fails |
| Business | Use AI for research and a first draft of the business case | Defend the case to a practitioner panel, including what you would not do |
| Every artefact | Ship it with a verification log: what was checked, how, and what failed | Talk a reviewer through the log and re-run one check live |
How a studio grade is made
Why the second test matters. In a randomised trial with high-school students, those who practised with an unguarded AI assistant scored 17% lower on a later exam taken without it. A version that gave hints instead of answers removed most of that harm. Bastani et al., PNAS 2025
Learn it first. Then let the agents work.
The trial above is why Year 1 is different. Students first build the understanding AI cannot give them, then use AI the way employers do.
Hints, not answers
AI explains and questions; students write, read and debug the code and the analysis themselves. Debugging and explanation carry the most weight in the no-AI tests.
Full agents, with logs
Students build with coding agents and workflow tools the way employers do, and keep a log of every check they ran on the output.
We teach the category, not the brand.
The best model this term may not be the best next term. Students learn one tool per category deeply, compare it with a second, and learn to choose. The stack is reviewed every term.
Six things you will remember doing.
Each one produces evidence an employer can check: a ledger, a deployed agent, a client’s sign-off, a recording of a defence.
First Rupee venture
Start a venture with a small seed and earn your first rupee from a real customer. Graded on process, not revenue.
Agent Lab
Ship a deployed agent each year, each with harder tools, stricter guardrails and a stronger evaluation suite.
SME consulting sprint
A paid sprint for a small business: map the process, automate what is worth automating, and hand it over working.
Bharat immersion
Time inside a business outside the metros, to see how money, risk and work really move in India.
Paid industry semester
A semester inside an employer on real work, [where the partner structure allows].
Demo Day
Present what you built to employers and investors, and defend it live. Ventures that survive get an incubator route.
A premium track, under your academic authority.
InfyAI delivers a quarter to two-fifths of the credits inside your regular, on-campus B.Tech, BBA or MBA. We bring the studios, the practitioners, the employer sprints and the evidence trail.