Prof Rod · Visiting facultyUniversities · Business schools · Companies

Guest lectures & workshops

Teach the agentic era
with the evidence attached.

Rod Rivera teaches how organizations change when intelligence becomes cheap and plentiful, and how to build and run the agent systems underneath. Sixteen sessions, adapted to your cohort, from the curriculum taught at ITAM, CDTM/TUM in Munich, LBS and LSE.

16 sessions · 2 tracks60 minutes to 6 hoursIn person or remote

Invitation

Bring a session to your program

Send the audience, the slot and the term. Rod replies within 48 hours with an outline written for your cohort.

Or read the full session menu

No fees on this page. A guest lecture and a teaching day are not the same ask, so they are scoped in conversation.

Why invite this particular outsider

What your program actually gets

Theory

The canon, audited, with the theorists credited

Coase and Williamson on why firms exist, Simon on bounded attention, Beer and Ashby on control, Penrose on growth, Nonaka on organizational memory. Each receives a verdict. Nothing here claims management theory is dead; the claim is that it is being re-sorted, and the room is shown the working.

Evidence

The failures are on the slides too

Anthropic's Project Vend put an agent in charge of a real shop and it lost money. Gartner expects over 40% of agentic-AI projects to be canceled by the end of 2027. Both are taught at full volume. A cohort that only hears the success cases has not been taught anything it can use.

Practice

A live example, not a hypothetical

Rod runs a media organization staffed by a layered agent workforce under written roles and deterministic gates, with one accountable human judging every output. Sessions draw on that operating record, including the parts that broke. The recursion is the point.

Logistics

No preparation burden on the host

Send the audience, the level and the slot. Rod adapts the session to your cohort, shares slides and worked material with participants afterwards, and leaves recording and reuse entirely to your institution's policy.

The point of the hour

What participants leave able to do

Management & business cohorts

Give a defended verdict on which management theories survive an agent workforce, and which dissolve
Set targets and read variance for work delegated to agents, using plan-versus-actual (Soll/Ist) thinking
Cost an agent operation, including the outputs that failed verification and returned nothing
Say where a human must remain accountable, and why that accountability cannot be delegated
Separate what has been demonstrated from what is being sold

Data science & applied AI cohorts

Read a token bill the way an analyst reads a cost center
Design a verification gate and state what it catches and what it misses
Diagnose a failure trace and name the layer that failed
Wire tools through the Model Context Protocol, and break them on purpose
Credential an agent to least privilege and decide what may run unattended

"A cohort that only hears the success cases has not been taught anything it can use on Monday."

Match it to your slot first

Choose a format

Any topic below runs in any of these. Pick the shape that fits your timetable, then pick the topics.

Format

What it is

Length

Guest lecture

One topic delivered to a full cohort, adapted to level. Slides and worked material shared afterwards.

Best forA single slot inside an existing course

60–90 min

Seminar

The same material with the room talking back: discussion, worked cases, and questions taken at length.

Best forSmaller cohorts and faculty groups

90 min

Hands-on workshop

Teaching plus a guided lab. Participants work at their own machines, on their own repositories where possible.

Best forTechnical and applied programs

Half day

Teaching day

The full arc: framing, instruments, and an afternoon design sprint or build-along. There are two: one strategy, one build.

Best forExecutive education and intensives

6 hours

Multi-session module

Topics sequenced into a running module across a term, assessed if your program requires it.

Best forA course that needs a whole strand

Five sessions or more

The menu · sixteen sessions

Pick the topics that fit your audience

Two tracks: one for management and business audiences, one for data science and applied analytics. Open any session for the full description, the outcomes and the audience fit. Topics combine naturally into half-day and full-day programs, and one session sits deliberately on the bridge between the tracks.

Track one

For business schools and management programs

MBA, EMBA, Master in Management, executive education, and strategy and organization courses.

The framing question of the discipline. What a firm looks like when headcount is no longer the unit of scale, why zero is an asymptote rather than a census, and what is still open versus already demonstrated. Grounded in cases, including the failures.

Participants leave able to

State what a firm is for when headcount stops being the unit of scale
Separate what has been demonstrated from what is being promised, case by case
Apply the coordination-cost argument to an organization they already know

Format: Lecture, extends to a half-day seminar
Audience: Business school audiences, management faculty and executive programs.

A structured audit of the canon against the agentic era. Taylor finally gets his fully-specifiable worker. Beer's Viable System Model turns out to describe agent organizations uncomfortably well. Span of control collapses into verification bandwidth, and organizational memory inverts when the firm forgets nothing but its workers forget everything. Each theory receives a verdict (validated, transmuted, bounded or dissolved), with the theorists credited by name. The audit was run three times, and seven of eleven verdicts moved.

Participants leave able to

Give a defended verdict on the theories they teach, or were taught
Name where span of control becomes verification bandwidth, and what that costs
Explain why organizational memory inverts under an agent workforce

Format: Lecture or seminar
Audience: MBA, EMBA and Master in Management cohorts; management faculty.

Built on the German Controlling tradition (Horváth, Küpper, Weber and Schäffer) and its central image of the navigator. German cost accounting spent a century establishing that there is no single true cost of a unit of output, and that steering on the wrong figure kills good businesses. That tradition arrives exactly when the first organizations appear whose workers are metered by the unit. Plan-versus-actual (Soll/Ist) thinking for delegated AI work, variance as a question rather than an accusation, and why the operator of an agent workforce is the Controlling function.

Participants leave able to

Set plan-versus-actual targets for delegated work and read variance as a question
Cost an output that failed verification: it spent tokens and review minutes and returned nothing usable
Locate the Controlling function inside an agent operation

Format: Seminar
Audience: Finance, controlling and operations audiences. No engineering background required.

What the role requires (delegation design, verification, cost governance), what it does not require, and how it differs from the CIO and CDO roles it will be confused with. Developed through invited engagements at TUM and LSE.

Participants leave able to

Describe what the mandate requires, and what it emphatically does not
Distinguish the role from the CIO and CDO seats it gets folded into
Draft the first ninety days of the mandate for a company they know

Format: Lecture
Audience: Executive education, EMBA and leadership programs.

This session takes the failure evidence at full volume and asks the management questions it raises: where agent economics actually break, what verification costs, and what separates the deployments that hold from the demos that do not. Works well in rooms that are already sceptical, and in rooms that are not sceptical enough.

Participants leave able to

Locate where agent economics break, using the published record rather than intuition
Price verification as a line item instead of an assumption
Tell a deployment that will hold from a demo that will not

Format: Case method, discussion-based
Audience: Case-teaching formats; MBA and executive cohorts.

The media organization Rod operates, where a layered agent workforce drafts, builds and checks the work under written roles and deterministic gates, and one accountable human judges every output. What the operating record actually shows. The recursion is the point: the organization that teaches this is run by it.

Participants leave able to

See what written agent roles, deterministic gates and a failure ledger look like in practice
Read an operating record and say where the human judgment sits
Take the artifacts away in a form they can copy

Format: Live tour
Audience: Any cohort. Works as an opener or as a closer.

Teams sketch an agentic operating model for a real business: which tasks are delegated, where verification sits, what the human retains. Closes with the failure evidence and a disciplined read on what is proven versus promised.

Participants leave able to

Leave holding a sketched operating model for a real business
Defend that sketch against the failure evidence
Hold a disciplined position on what is proven versus what is promised

Format: Executive-education format, with an afternoon design sprint
Audience: Executive education, EMBA, faculty development and leadership teams.

Track two

For data science, analytics and applied AI programs

Master’s-level data science, business analytics, applied AI tracks, technical bootcamps and career-transition programs. Semi-technical is welcome: sessions assume comfort with data, not with production engineering.

The anatomy of an agent that runs continuously on infrastructure you own: the loop, memory, tools, and the harness around them. Uses a real task, run live, with the failure modes left in. The right opening session for a program new to the material.

Participants leave able to

Name the four parts of an agent (loop, memory, tools, harness) and what each is for
Say what separates a chatbot from a colleague-shaped system
Ask the two questions that decide whether it was worth it: what did it cost, and how would you know if it went wrong

Format: Lecture with live demonstration
Audience: Mixed technical and non-technical. No prior agent experience assumed.

The engineering layer underneath agent reliability: state machines, validators, gates, exception routing, and why capable models still fail in production. Covers the ladder from asking for evidence, to goal conditions, to deterministic gates, and where each rung stops being safe. The half-day version adds a guided lab diagnosing real failure traces.

Participants leave able to

Diagnose a failure trace and name the layer that failed: model, harness or brief
Build validators, gates and exception routing around a capable-but-fallible model
Design work so a session cannot claim it is finished until a check has actually passed

Format: Lecture, with a hands-on lab option
Audience: Technically capable operators and engineering teams already using agents and hitting the limits.

Practical patterns for budget-aware systems, and the analytical skill underneath them: reading a token bill the way an analyst reads a cost center.

Participants leave able to

Read a token bill the way an analyst reads a cost center
Budget a context window and defend the trade-offs made to fit it
Recognize the patterns that quietly multiply cost at production scale

Format: Lecture or seminar
Audience: Master’s-level data science, business analytics and applied AI tracks.

The five tool-calling paradigms, the failure surface of each, and the Model Context Protocol as the emerging standard. The hands-on version has participants wire tool calls themselves, and then break them on purpose.

Participants leave able to

Choose a tool-calling paradigm for a given task and justify the choice
Wire a tool through the Model Context Protocol
Break one deliberately and read what the failure tells them

Format: Lecture, with a hands-on lab option
Audience: Applied AI and engineering tracks. The lab needs participants at their own machines.

Retrieval, structured stores, and the organizational reading that follows: the system's memory is the organization's memory.

Participants leave able to

Choose between retrieval and structured stores for a stated workload
Anticipate the memory failures that never show up in a demo
Read the system's memory as the organization's memory

Format: Lecture or seminar
Audience: Applied AI, data science and engineering tracks.

Recursive context patterns, when retrieval helps and when it does damage that is hard to see, and evaluation approaches that actually catch it.

Participants leave able to

Build recursive context patterns that hold at depth
Detect retrieval poisoning with evaluations designed to catch it
Say when retrieval is simply the wrong instrument

Format: Lecture
Audience: Master’s-level data science and applied AI tracks.

Verification gates, behavioral checks versus presence checks, instrumenting agent work from birth, and plan-versus-actual thinking applied to AI output. A green compile is necessary and never sufficient. Verification is the new quality assurance, and this session is where the two tracks meet.

Participants leave able to

Write a verification gate and state precisely what it catches and what it misses
Tell a behavioral check from a presence check
Instrument delegated work from birth so variance can be read later

Format: Lecture or workshop
Audience: Analytics and management audiences alike: the most transferable session on the menu.

Sober and checklist-backed, free of both hype and doom.

Participants leave able to

Credential an agent to least privilege, and prove that it is
Defend against prompt injection and untrusted data inside the loop
Run the checklist that decides whether something may run unattended

Format: Lecture or workshop
Audience: Engineering, applied AI and security-adjacent tracks.

The hands-on teaching day, run on infrastructure patterns participants can reproduce afterwards without a vendor relationship.

Participants leave able to

Ship a working agent on patterns they can reproduce at home
Put a harness, gates and observability around it
Leave with the repository and the checklist

Format: Build-along; participants work at their own machines
Audience: Assumes basic scripting comfort. Everything else is taught in the room.

Where this material is taught

ITAM

Professor of the Practice

CDTM / TU Munich

Courses in Munich, in the school where Rod himself trained

LBS & LSE

Workshops and invited engagements

Nebius Academy

A five-week AI engineering course

NVIDIA

Tutorials and content partnership

Who teaches this

Rod Rivera

Rod builds and studies organizations where one human directs a workforce of AI agents. He teaches the material below because he runs it first: every session draws on an operating record rather than a forecast, and the failures are taught at the same volume as the wins.

The classroom

Professor of the Practice

ITAM, with teaching at CDTM/TUM in Munich, workshops at LBS and LSE, and courses at Nebius Academy.

The machinery

15+ years in production ML

Machine learning systems at Alibaba Cloud, Huawei and Samsung.

The research

PhD · 20+ publications

Machine learning research across IEEE, SIGKDD and other peer-reviewed venues.

The book

An agent from scratch

Twenty-one chapters in Python. Each derives the math, runs on Colab and measures a real model.

More about Rod

For whoever has to plan this

The practical details

Group size: From a seminar of twelve to a full lecture hall. Hands-on labs work best under forty.
In person or remote: London-based, and regularly in Munich and Mexico City. Remote delivery is fully supported and is not treated as the lesser version.
Languages: English, Spanish and German.
Materials: Slides, worked examples and the code are shared with participants after the session.
Recording: Recorded, distributed and reused entirely according to your institution's policy.
Adaptation: Every session is rewritten to the cohort before delivery. Send the audience and the course it sits inside.
Lead time: Six weeks is comfortable. Less has been done, and more than once.
Fees: Scoped per engagement and discussed directly. A guest lecture and a teaching day are different asks, so there is no single figure.

Bringing this to your program

Tell Rod which room this is for.

The topics that fit your audience, the length you have in mind, and the term you are planning for. Rod replies within 48 hours with a session outline adapted to your program and the practical details, fees included and scoped to the engagement.

Or email [email protected]

Companies: the same menu adapts to team formats. Write with your team's context and Rod will scope it.