TMBA Assignment Learning Bite · Flinders

The Compass
and the Tribe.

A modern fable about Agentic AI in marketing.

The fable of the Compass: the dragon of churn, the conjured scrolls, the test, and the tribe lit like little suns churn The fable of the Compass

What awaits you on the journey.

You will start with the big picture, follow a fable that shows Agentic AI at work, build a real agent brief yourself, test what you learned, and leave us a quick note.

Your Learning Objective

By the end of this Learning Bite, you will be able to explain how an Agentic AI campaign differs from traditional and GenAI-assisted approaches and articulate the three mechanisms that make it work: Generate, Test, Adapt.

Chapter I · Bird's Eye

From tools that wait for orders
to systems that act on their own.

The Problem – John Wanamaker, ~1900
"Half the money I spend on advertising is wasted. The trouble is, I don't know which half."

In 2026, 51% of marketing professionals report that their teams frequently run generic campaigns. More than a century after Wanamaker’s famous quote, this points to a familiar tension: While marketing has become more measurable and automated, many teams still struggle to determine which message resonates with which person. Agent-based AI offers a new breakthrough: campaigns that continuously learn from individual responses.

Salesforce State of Marketing 2026 · 10th Edition · 4,500 marketers worldwide
Stage 1 · Traditional

Broadcast Marketing

One message. One audience. One channel. Every decision needs to be approved before launch, results are revealed only weeks later, and not a single real person actually reached.

Stage 2 · GenAI

Generative AI

Content gets made faster. The copy gets better. But a human is still making every call. The marketer prompts, the AI writes. Wanamaker's problem gets smaller, but it doesn't go away.

Stage 3 · Today

Agentic AI

The marketer defines the goal. The agent generates, tests, and adapts on its own, inside the guardrails a human sets, and adapts continiously.

Definition
Agentic AI

"An AI agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors. "
Russell & Norvig, Artificial Intelligence: A Modern Approach, 4th ed. (Pearson, 2020)

I.

It perceives its environment

An agentic system pulls signal from your CRM, web analytics, conversation data, and inboxes. It does it continuously so that it's not needed to prompt it each time.

II.

It makes decisions

It chooses what to do next within the guardrails you set: which segment to target, which channel to use, when to escalate to a human.

III.

It takes actions autonomously

It sends messages, tests and learns by itself.

75%
of marketing organizations use at least one form of AI
51%
of marketers say their teams frequently run generic campaigns
+20%
Marketing ROI increase reported by marketers who successfully deploy AI in their campaigns
1.9×
High-performing marketing teams are 1.9x more likely to have AI agents than underperformers
Source: Salesforce, State of Marketing, 10th Edition, 2026
Watch Before You Begin

You have just seen the numbers: most marketers now use AI, yet most still broadcast generic campaigns. So what does an agent that actually closes Wanamaker's gap look like in motion? Take two minutes to watch a short video.

▶ Agentic AI in Marketing

Step into the Fable

A brief introduction from Darina.

Video could not be loaded
If you can't see the media, please reach out to Jochen Schuba.

Before you meet the stranger, give your Agentic AI a name (make it yours).

Throughout this story you will follow an Agentic AI as it helps a struggling tribe. By default we call it Claude.

Chapter II · Deep Dive

The fable of Sofia,
Janus, and the Compass.

Every scene maps to a real capability of an agentic marketing system. Watch for them. What follows is the condensed version. For those who want every detail, the supplementary full story waits at the end of this chapter.

🐉 The Dragon · Churn

Deep in the valley, surrounded by ancient forests and rolling hills, lives an ancient tribe..

It lives, trades, celebrates. And slowly loses its members.

Sofia is a member of the tribe's Communications Council. Her job, along with her team, is to ensure that every tribe member feels seen, heard, and valued by reaching the right person at the right time with the right message.

Currently, they are facing a situation where members are leaving faster than new ones are joining. To put a stop to this and regain their enthusiasm, the team spent weeks developing two campaigns, discussing messages and channels, and pulling the data team away from an important analysis for ensuring the food supply. After all that, the campaigns finally go live. Two days later, they hear:

"843 members left last month. A new record. We doubt the positive impact of your current measures."

Two days. The ink on the campaigns has barely dried, and the verdict is already in.

⚔️ The Stranger · A Compass Appears

While the team sits around the campfire, a stranger appears.

While the team sits around the campfire, still waiting for the first results and already thinking about the next step, a stranger appears.

Old, sharp-eyed and wearing a worn leather bag. His name is Claude.

He places something on the table. It looks like a compass, but instead of pointing north, it points in all directions at once. Hundreds of tiny lights, moving.

"843 members left for 843 different reasons. Two campaigns were your only response. This dragon isn't stronger than you, but it is very fast. And it knows every tribe member personally. You don't, you can only guess how to divide them into groups."

"What is that thing?" Sofia asks, looking at the object on the table. Claude smiles. "This is how you'll get to know them, too."

"Every light" Claude says softly, "is one of your tribe members. And this will help you reach every single one. Not as a group. But individually."

I. Generate

Not two messages per 3 weeks. Hundreds. Adapted to every individual member and drafted in minutes.

Traditional marketing
2 messages in 3 weeks

II. Test

Not one A/B test after a month. 312 simultaneous micro-tests. First results within hours. Variants compete with each other and the best survive.

Traditional A/B testing
1 test per quarter

III. Adapt

No review meeting. The compass decides what works and do it overnight, within the guardrails Sofia set.

Traditional optimization
Quarterly review cycles
⚡ Generate · The Scrolls Multiply

"Not two messages. Hundreds."

He runs his hand over the compass, and suddenly tiny scrolls appear on the table. First two, then four, more and more, endlessly.

"Different words for the young hunter who hasn't opened your app in weeks. Different words for the grandmother who receives your monthly scroll but never replies. And yet more words for the family in the northern hills who recently lost their landline."

"They all appear within minutes. Not because it is guessing, but because it can learn from patterns in past replies, clicks, and ignored messages, if that data is made available to it. It speaks the language of your tribe because it has learned the language."

❦ Learning 1 The generative layer is based on large language models that have been pre-trained on vast amounts of human-written text, enabling them to generate context-sensitive content at scale based on prompts and examples (Brown et al., 2020). When combined with individual behavioral and interaction data, these models can be used to create tailored messages for specific members of a target audience without having to manually compose each message (Salesforce, 2026).

"But how do you know which message works?" Sofia asks.

🏹 Test · Not Testing. Learning.

"Not just testing. Learning. There is a difference."

"You've created two messages, spent weeks of work before launch, and now you're waiting weeks for results. This tool sends them all out at once. To small, data-driven, and carefully selected groups. And within a few hours" he picked up three scrolls and set them aside gently, "it already knows. These three work; the conversion rate is 20%. The others" he quietly swept the rest off the table, "don't."

"You're testing everything at once?" Sofia asks.

"Not just testing. And not me. The Compass does it. It tries things out and learns as it goes."

❦ Learning 2 The "multi-armed bandit" approach maintains a continuous balance between exploration and exploitation: it tests multiple message variants while gradually shifting focus to those that perform best. Unlike traditional A/B tests, it does not have to wait for a predetermined test to end before reallocating attention. In practice, this means starting with small audience segments, evaluating real-world responses, and scaling the variants that work, with fewer manual decisions required between rounds (Sutton & Barto, 2018). This is what the Compass does while Sofia sleeps.
🛡️ Doubts · Janus Speaks

"HOW IS SOMETHING LIKE THAT SUPPOSED TO WORK??"

Unease spreads through the tent. Janus, one member of the group, furrows his brow. Tim is staring at the wall, looking confused.

"HOW IS SOMETHING LIKE THAT SUPPOSED TO WORK?? Imagine, we needed to talk to the horse messenger, we negotiated contract conditions, we clarified our budget internally! And you're telling us a machine can substitute all of this? All this effort, all of our knowledge about our tribe members? You're a stupid fool!"

The tent is loud now. Claude stands still. He doesn't raise his voice.

"Janus. How long have you been with this tribe?"

"Seventeen years."

"And in seventeen years, how many campaigns have you run?"

"Hundreds."

"Then you are exactly the person this compass needs."

The room goes quiet.

"This does not replace your knowledge of the tribe. It cannot negotiate with the horse messenger as one of your channels, you do that. It cannot decide what your tribe stands for, you do that. It cannot set the boundaries of what is acceptable, you do that. What it can do is watch all the tribe members simultaneously and adjust every message every hour. That is not a question of expertise, think about all the time you needed for these two campaigns, it is a question of scale."

"You set the rules. You define the budget it cannot exceed. You decide which tribe members it can and cannot contact. And then, within those boundaries which YOU set, it acts, based on all the information which you shared with it. Without this data, this compass won't work."

❦ Learning 3 "Human-in-the-loop" means that an AI agent operates within the framework of goals, constraints, and control mechanisms defined by humans. The agent can perform routine tasks (Russell & Norvig, 2020), but decisions that exceed its authority, safety assessment, or risk threshold are escalated to a human expert. The agent executes; the human governs (Shneiderman, 2020).

Janus steps forward reluctantly and transfers the information to the Compass. Seventeen years of knowledge. Julia, another member, stands up and gives a detailed overview of the channels.

"Now let's watch. Go home and get some sleep. The Compass doesn't need that. Let's meet tomorrow at 9 a.m. and see what's happened."

You can watch the compass work.

The Compass sits on the table. Sofia has set the guardrails. Janus, reluctantly, has shared seventeen years of knowledge. There is nothing left for the team to do tonight. Click below to leave, sleep, and let the night unfold.

0
micro-tests launched
0
winners detected
0%
budget reallocated
Guardrails received from Sofia.
Goal: reduce churn among at-risk members.
The Compass is ready. Awaiting your command.
🏆 Adapt · Morning After

Sofia is already there when the others arrive. She hasn't slept much.

In the morning, the first people enter the room. The compass is still on the table, glowing softly. But something looks different, some stars are shining a little brighter than last evening, a handful of them like little suns.

Claude is the last to come in. He sets down his bag, pours himself some tea, and sits down. He says nothing, but just looks at Sofia.

"It looks like it's still working" says Sofia.

"It has been working all night" replies Claude. "Those brighter stars represent tribe members the compass has reached. Based on everything we gave it yesterday: their history, their behavior, their interests. And we can see that these lights have opened the message."

Janus asks skeptically, "How many messages has it sent?"

"312" says Claude.

Janus laughs briefly and dryly. "312. We have hundreds of thousands of tribe members. That's nothing."

"Last month, your team sent two types of messages" says Claude. "Last night we sent 312, each one different, each one selected. Tonight it will send more. The next night, even more. And each time, the compass knows a little more than it did last time."

"And the ones that didn't work?"

"The Compass has already stopped them and automatically redirected the budget to what did work. Even before you came in this morning."

❦ Learning 4 The adapt loop is a core capability of an agentic system: the system observes outcomes, updates subsequent actions, and reallocates attention or resources toward what performs better. Unlike a scheduled campaign, it does not simply run to completion; it keeps learning between cycles, within human-defined goals and constraints (Russell & Norvig, 2020; Salesforce, 2026).

"So it never ends" Sofia says slowly.

"It never ends. Your old campaigns had a beginning and an end, along with all that preparation time. This one has a beginning, and then it just keeps getting better and better."

Silence follows. Janus walks over to the table. He looks at the compass, at the stars, at the numbers Sofia read aloud. He stands there for a moment that feels longer than it is.

"It took everything we knew" Sofia says quietly, "and turned it into something that never sleeps."
Janus nods slowly. "Same goal, same budget, same members. But last night, we actually reached them."

❦   The compass glows on.   ❦

Supplementary reading: the complete version · ~10 minutes · every scene, every detail

❦ The Complete Fable – Unabridged

The Tribe

Deep in the valley, surrounded by ancient forests and rolling hills, lives an ancient tribe. The tribe consists of hunters and farmers, artisans and families who live together, trade, and celebrate festivals. Their bonds run deep. But the tribe is large, and it keeps growing more complex.

Sofia is a member of the tribe's Communications Council. Her job, along with her team, is to ensure that every tribe member feels seen, heard, and valued by reaching the right person at the right time with the right message, a task that sounds straightforward until you are responsible for hundreds of thousands of people.

The Dragon Stirs

Today, the council is facing a growing threat: members are leaving faster than new ones are joining. The team comes together to find answers. Should they offer something special to those at risk of leaving? Is there something missing in what the tribe provides? Have other tribes been more persuasive?

The discussion goes on for hours. The bonfire crackles. Ideas pile up on the table. After a long session, two campaigns take shape. The first targets app users with a message about a new connectivity feature. The second reaches non-app members with a combined offer for mobile and fixed line access.

They send selection requests to the data team. Group 1: members who logged in at least three times in the last two months, no advertising objectors. Group 2: members without app activity in three months and without a mobile contract.

Five days pass. The data team delivers, having had to push back a critical food supply analysis to make it happen. Three weeks later, both campaigns finally go live, and for the first time in a while, the team allows itself a moment of relief, knowing that results will take at least another two weeks to come in.

While the team is waiting, the tribe leadership sends an alarming message: last month, 843 members left, a new record, and the verdict is already in before the campaigns have even had a chance to prove themselves: we doubt the positive impact of your current measures.

The Stranger

The team sits around the campfire, still waiting for the first results and already thinking about the next step, when a stranger appears at the entrance. Nobody saw him arrive. Old, sharp-eyed, with a slight smile on his face. He carries no sword, no shield. Just a worn leather bag.

"Hi, I'm Claude. I was traveling through your valley and couldn't help but overhear" he says calmly. "843 members in one month. And your campaign went live when, exactly?" "Just two days ago" someone mutters.

"And how many different messages did you send?" "Two" Sofia says. "App users and non-app users."

Claude listens. His eyes sharpen. He opens his bag and places something on the table. It doesn't look like anything they have seen before.

"843 members left for 843 different reasons. Two campaigns were your only response. This dragon isn't stronger than you, but it is very fast. And it knows every tribe member personally. You don't, you can only guess how to divide them into groups."

"What is that thing?" Sofia asks. Claude smiles. "This is how you'll get to know them too."

The object glows softly, a compass that points everywhere at once. Hundreds of tiny lights, moving. "Every light is one of your tribe members. And this will help you reach every single one. Not as a group. But individually."

Generate

"The first thing this compass does is generate. Not two messages. Hundreds." He runs his hand over the compass and suddenly tiny scrolls appear on the table. First two, then four, more and more, endlessly.

"Different words for the young hunter who hasn't opened your app in weeks. Different words for the grandmother who receives your monthly scroll but never replies. And yet more words for the family in the northern hills who recently lost their landline."

"They all appear within minutes. Not because it is guessing, but because it can learn from patterns in past replies, clicks, and ignored messages, if that data is made available to it. It speaks the language of your tribe because it has learned the language."

❦ Learning 1The generative layer is based on large language models that have been pre-trained on vast amounts of human-written text, enabling them to generate context-sensitive content at scale based on prompts and examples (Brown et al., 2020). When combined with individual behavioral and interaction data, these models can be used to create tailored messages for specific members of a target audience without having to manually compose each message (Salesforce, 2026).

"But how do you know which message works?" Sofia asks.

Test

"You don't" Claude says simply. "And neither does it, at first. But here is the difference." He spreads a hundred tiny scrolls across the table.

"You created two messages and now you are waiting weeks for results. This tool sends all of them simultaneously. To small, data-driven, and carefully selected groups. And within a few hours" he picked up three scrolls and set them aside gently, "it already knows. These three work, conversion rate of 20%. The others" he quietly swept the rest off the table, "don't."

"You're testing everything at once?" Sofia says. "Not just testing. And not me. The Compass does it. It tries things out and learns as it goes."

❦ Learning 2The "multi-armed bandit" approach maintains a continuous balance between exploration and exploitation: it tests multiple message variants while gradually shifting focus to those that perform best. Unlike traditional A/B tests, it does not have to wait for a predetermined test to end before reallocating attention. In practice, this means starting with small audience segments, evaluating real-world responses, and scaling the variants that work, with fewer manual decisions required between rounds (Sutton & Barto, 2018). This is what the Compass does while Sofia sleeps.

Doubts

Unease spreads through the tent. Janus, one member of the group, furrows his brow. Tim is staring at the wall, looking confused.

"HOW IS SOMETHING LIKE THAT SUPPOSED TO WORK?? Imagine, we needed to talk to the horse messenger, we negotiated contract conditions, we clarified our budget internally! And you're telling us a machine can substitute all of this? All this effort, all of our knowledge about our tribe members? You're a stupid fool!"

Claude stands still. He doesn't raise his voice. "Janus. How long have you been with this tribe?" "Seventeen years." "Then you are exactly the person this compass needs."

"This does not replace your knowledge of the tribe. It cannot negotiate with the horse messenger as one of your channels, you do that. It cannot decide what your tribe stands for, you do that. It cannot set the boundaries of what is acceptable, you do that. What it can do is watch all the tribe members simultaneously and adjust every message every hour. That is not a question of expertise, think about all the time you needed for these two campaigns, it is a question of scale."

"You set the rules. You define the budget it cannot exceed. You decide which tribe members it can and cannot contact. And then, within those boundaries which YOU set, it acts, based on all the information which you shared with it. Without this data, this compass won't work."

❦ Learning 3"Human-in-the-loop" means that an AI agent operates within the framework of goals, constraints, and control mechanisms defined by humans. The agent can perform routine tasks (Russell & Norvig, 2020), but decisions that exceed its authority, safety assessment, or risk threshold are escalated to a human expert. The agent executes; the human governs (Shneiderman, 2020).

Janus steps forward reluctantly and transfers the information to the Compass. Seventeen years of knowledge. Julia, another member, stands up and gives a detailed overview of the channels. Claude steps back. The compass glows, not static. Moving.

"Now let's watch. Go home and get some sleep. The Compass doesn't need that. Let's meet tomorrow at 9 a.m. and see what's happened."

Adapt - The Morning After

In the morning, the first people enter the room. The compass is still on the table, glowing softly. But something looks different, some stars are shining a little brighter than last evening, a handful of them like little suns.

Claude is the last to come in. He sets down his bag, pours himself some tea, and sits down. He says nothing, but just looks at Sofia.

"It looks like it's still working" says Sofia.

"It has been working all night" replies Claude. "Those brighter stars represent tribe members the compass has reached. Based on everything we gave it yesterday: their history, their behavior, their interests. And we can see that these lights have opened the message."

Janus asks skeptically, "How many messages has it sent?" "312" says Claude.

Janus laughs briefly and dryly. "312. We have hundreds of thousands of members. That is nothing." "Last month your team sent two types of messages. Last night we sent 312, each one different, each one selected. Tonight it will send more. The next night, even more. And each time, the compass knows a little more than it did last time."

"And the ones that didn't work?" "The Compass has already stopped them and automatically redirected the budget to what did work. Even before you came in this morning."

"But how does it decide when to stop? When to move the budget? That requires experience." Claude leans forward. "You are right. But not the kind that needs to sleep. Every hour last night, the compass looked at what was happening. And based on what it saw, it adjusted. Stopped what wasn't working. Doubled what was. Not once. Continuously."

❦ Learning 4The adapt loop is a core capability of an agentic system: the system observes outcomes, updates subsequent actions, and reallocates attention or resources toward what performs better. Unlike a scheduled campaign, it does not simply run to completion; it keeps learning between cycles, within human-defined goals and constraints (Russell & Norvig, 2020; Salesforce, 2026).

"So it never ends" Sofia says slowly. "It never ends. Your old campaigns had a beginning and an end, along with all that preparation time. This one has a beginning, and then it just keeps getting better and better."

Silence follows. Janus walks over to the table. He looks at the compass, at the stars, at the numbers Sofia read aloud. He stands there for a moment that feels longer than it is.

"It took everything we knew" Sofia says quietly, "and turned it into something that never sleeps."
Janus nods slowly. "Same goal, same budget, same members. But last night, we actually reached them."

❦   The compass glows on.   ❦

Chapter III · Activity

Now you hold
the compass.

You have seen what the compass can do. Now hold it yourself.

Your Quest

Draft an agent brief for a product you actually care about.

Set the guardrails on the left. Watch the brief on the right come to life. Each rule you select becomes a line in the prompt.That's exactly the way Sofia handed instructions to the Compass. When the brief reads the way you want, copy it into an LLM and see what your "compass" produces.

Step 1 · Sofia's Decisions

Set the guardrails.

Click each rule you want the agent to follow. Each click writes a line into the brief on the right.

0 / 5 guardrails set
Step 2 · The Agent Brief
// Agent brief – guardrail-bound

You are a marketing agent for [a sustainable coffee subscription].

GOAL: Reduce churn among [customers who haven't ordered in 30+ days] by [15% over the next month].

GENERATE: Draft 5 distinct re-engagement message variants that vary in tone (warm / playful / informative / curious / urgent) and in offer (no offer / 15% off / free shipping / new product reveal / loyalty perk).

TEST: For each variant, predict which customer segment it would resonate with most, and what signal would tell us it's working within 48 hours.

GUARDRAILS:

BRAND VOICE: [ not set - click the chip on the left ]

COMMERCIAL LIMIT: [ not set ]

DATA PRIVACY: [ not set ]

ESCALATION: [ not set ]

SUCCESS METRIC: [ not set ]

Output as a table.

Step 1
Generate

Paste the brief into an LLM. Read all 5 variants. Which one surprises you?

Step 2
Test

If you ran all 5 tomorrow, which two would you actually launch and why?

Step 3
Adapt

What guardrail did you forget to set? Add it to your brief and run it again. What changed?

Share Your Observation (optional)

Head over to our Canvas discussion and post one thing your "compass" produced that surprised you (a variant you would never have written, a guardrail you almost forgot, or a result you did not expect). Then read what your classmates discovered. That is how the tribe learns together.

❦   Share in the Canvas discussion   ↗
Chapter IV · The Test

Your take aways
from the fable.

Question I

In the context of the evolving marketing landscape, what primary characteristic distinguishes Agentic AI from Generative AI?

Unlike Generative AI, which produces content only when prompted by a human, Agentic AI continuously perceives its environment and takes autonomous actions to achieve a defined goal, without needing a hand on the steering wheel at every step.
Question II

What does an agentic system use as percepts to understand its environment?

An agentic system reads live signals from CRM data, web analytics, conversation history, and inboxes on a continuous basis, not manual reports, not random variants, and not just what a human types into a prompt box.
Question III

According to the Multi-armed bandit approach mentioned in the learning material, how does testing differ from traditional A/B methods?

The multi-armed bandit approach runs hundreds of micro-tests simultaneously, exploiting what works in real time while still exploring new options - unlike traditional A/B testing, which runs one test, waits weeks for results, and requires a human decision before moving to the next round.
Question IV

What is the primary role of the human marketer within an agentic system, as illustrated by the Human-in-the-Loop principle?

In a Human-in-the-Loop model, the human marketer's job is to set the rules the agent operates within the budget ceiling, the brand voice, the escalation triggers, while the agent handles execution at scale. The human governs; the agent acts.
Question V

Which of these is least appropriate as a guardrail in an agentic marketing brief?

A guardrail is meant to constrain the agent's behavior and protect the brand. "Maximize click-through rate at any cost" does the opposite - it removes all limits and could drive the agent toward manipulative or brand-damaging tactics. Every other option sets a clear boundary; this one eliminates them.
YOUR SCROLL
0 / 5

Closing.

The Last Bite

Sergiu wraps up the key learnings.

Video could not be loaded
If you can't see the media, please reach out to Jochen Schuba.
Chapter V · Feedback

Feedback

Form not loading? Open it in a new tab ↗

Appendix · Further Reading

For those who want
to dive deeper.

i.
State of Marketing 2026: Foreword
Bobby Jania · Salesforce · 2026 · 2-minute video
A brief industry perspective on the transition from one-way broadcast campaigns to real-time, data-driven customer dialogues.
Video
ii.
State of Marketing Report – 10th Edition
Salesforce · 2026
Industry benchmarks on AI adoption, the struggle with generic campaigns, and the shift toward data-driven, real-time dialogues.
Report
iii.
Agentic Workflows
IBM Think
The technical mechanics behind the magic: understanding multi-step reasoning, iterative loops, and how agents evaluate their own work.
Guide
iv.
Reinventing Marketing Workflows with Agentic AI
McKinsey & Company
A deep dive into process redesign: how to break down traditional campaign silos and transition to automated, agent-driven pipelines.
Report
v.
AI Agents in Marketing
IBM Think
Practical use cases exploring the leap from basic generative tools to systems capable of dynamic, personalized, and autonomous actions.
Article
vi.
Generative Agents: Interactive Simulacra of Human Behavior
Park et al. · Stanford University & Google Research · 2023
Foundational research on the architecture of autonomous agents, detailing how AI systems use memory, reflection, and planning to continuously adapt and simulate human-like decision-making.
Academic
vii.
Intelligent Agents (Chapter 2)
Russell & Norvig · University of California, Berkeley
The complete foundational text exploring the nature of agents, operational environments, and the structural design of autonomous perception–action loops.
Textbook
About · Flinders Team

The team behind
the compass.

Johannes Amann
If you can't see the media, please reach out to Jochen Schuba
Johannes Amann
LinkedIn
Darina Istomina
If you can't see the media, please reach out to Jochen Schuba
Darina Istomina
LinkedIn
Sergiu Revitea
If you can't see the media, please reach out to Jochen Schuba
Sergiu Revitea
LinkedIn
Jochen Schuba
If you can't see the media, please reach out to Jochen Schuba
Jochen Schuba
LinkedIn
Luka Wingfield
If you can't see the media, please reach out to Jochen Schuba
Luka Wingfield
LinkedIn
Sources
References
  • Brown et al. (2020). Language Models are Few-Shot Learners. NeurIPS.
  • Russell, S. & Norvig, P. (2020). Artificial Intelligence: A Modern Approach. 4th ed. Pearson.
  • Sutton, R.S. & Barto, A.G. (2018). Reinforcement Learning: An Introduction. MIT Press.
  • Shneiderman, B. (2020). Human-centered artificial intelligence: Reliable, safe & trustworthy.
  • Salesforce (2026). State of Marketing. 10th Edition.
  • Agentic AI in Marketing - A Two-Minute Introduction (2026). Course-produced introduction video, hosted on Vidyard.
AI Disclaimer
Use of Artificial Intelligence

This Learning Bite is the original work of the listed group members. While Artificial Intelligence tools were used to support initial research, suggest structural outlines, and assist with the technical implementation of interactive elements, human oversight drove the entire project.


All AI-generated output was critically reviewed, substantially rewritten, and strictly edited. The overarching editorial direction, the tribal narrative, and the story about the characters of Sofia and Janus represent original human authorship.

Course Context
Assignment Information

Learning Bites Group Assignment · Agentic AI for Marketing
TMBA · Marketing and Sales in a Digital World · ESMT Berlin


Topic approved by course instructor. This page is intended solely for educational use within the course and does not represent the views of any commercial entity.

Image Credits
Visuals & Design

All illustrations on this page are original SVG graphics created for this assignment. The parchment aesthetic and medieval typography were designed to serve the tribal narrative framing of the learning content.