
Explore how AI agents differ from rule-based bots and RPAs, combining perception, memory, planning, and autonomous decision making to adapt, learn, and use tools in healthcare.
Bots, macros and AI agents side by side: what each works from, how aware it is of context, memory, learning, tools and communication. Then the analogy to remember (vending machine, assistant with a script, junior colleague), why agents became practical now, and an honest look at what still goes wrong and when a rule or macro is still the better choice.
Our first hands-on session. Meet Brewline Coffee, a fictional coffee company, and its Monday morning inbox of 12 emails. We give the same inbox to a rule-based bot and an RPA macro, score the results, and see exactly where rules break and where macros fail silently. In Part 2, an AI agent takes on the same inbox. Download the workbook (Brewline_Monday_Inbox.xlsx) to follow along, edit the keyword rules yourself, and decide where you would send each email.
The AI agent takes on the same 12 emails, and everything you see is its real output. Watch it read between the lines, look things up and hand the right decisions to people. Then see what happens when it takes a vague instruction at its word, and how one clear refund policy cut the emails handed back to humans from six to three. The workbook (Brewline_Monday_Inbox.xlsx) contains both real agent runs.
The rule-based bot, the RPA macro and the AI agent, side by side on the same inbox, and every row of the comparison chart from the lecture seen in practice. Then the leadership takeaway: when to use rules, when an agent earns its place, and where a human must decide. Use the one-page Agent or Automation checklist to test one of your own recurring tasks.
Explore how ai agents augment human capability across healthcare and enterprise, from diagnostic copilot and care coordination to email triage and meeting prep, including multimodal interaction.
Explore how planning agent, habit tracker agent, and life coach agent empower personal life management, plus multimodal agents in medicine, smart homes, and cross-channel support.
Hands-on practice for the examples tour. Brewline Coffee, our fictional company with 14 cafes and around 120 staff, asks where else agents could help. We give an AI agent twelve real-life processes and the pattern library from the lecture, and it scores each one on value, risk and readiness, with a verdict: pilot now, pilot with human approval, not yet, or do not use an agent. Then we look critically at its recommendations, including its worst idea. Everything on screen is real agent output. Attached: the Agent Use Case Canvas to score five processes from your own organisation, and the full Brewline opportunity scan.
Define how perception, memory, decision making, and action shape modern AI agents, using tokenization, embeddings, feature extraction, and memory with vector databases to connect inputs with past knowledge.
Hands-on practice for the lecture on core components. We open up the Brewline Care Agent from Hands-on 1 and give it Sophie's difficult "Thanks a lot." email. The agent shows its work in four stages: perception, memory, decision-making and action. Then we take one component away at a time (no memory, keyword-only perception, no decision rules) and watch what breaks. Everything on screen is real agent output. Attached: the Agent Anatomy one-pager with the four questions to ask about any agent demo.
AI agents use external tools to retrieve real-time data, manipulate software, and make dynamic decisions—via search, calculation, APIs and plugins, and code execution—while addressing latency, safety, and transparency.
How tools are connected in practice and what they risk. A travel agent chains four tools in a row. Developer frameworks explained with the Brewline Care Agent: describe each tool once, list what the agent may use, feed results back. MCP as the universal plug. Then the risks (slow, failing, powerful tools, prompt injection) and the four decisions that belong to leaders.
Hands-on practice for the lecture on tool use, Part 1 of 2. The Brewline Care Agent from Hands-on 1 gets seven real tools: order look-up, parcel tracking, carrier news, a calculator, warehouse tasks, a hand-over to humans, and email drafts (but no sending and no refunds). We watch it handle five customer emails with 24 logged tool calls, including a tracking service that times out, and see how it decides when a tool is needed and when not to act. Everything on screen is the agent's real tool log. Attached: the full Brewline tool log as a workbook.
Hands-on practice for the lecture on tool use, Part 2 of 2. Mina wants to cancel a prepaid year and get the unused months back: we see why agents need a calculator for every money calculation, and why a 76 euro refund above the agent's limit goes to a human. Then we score our real run on the lecture's four considerations (latency, failure modes, security, transparency), including the moment the agent admitted bending one of its rules, and look at MCP, the standard plug that connects agents to business systems. Attached: the Tool Permission Checklist with five questions to ask about every tool you give an agent.
A quick, illustrated tour of the newer kinds of tools beyond the four classic families: computer use (from Adept in 2022 to Anthropic, OpenAI Operator and Perplexity Comet), other agents, skills, memory, asking a human, and the exotic ones: agents that pay, agents with a body, and agents in the lab. Ends with the one question a leader should ask about every tool: does it only read, or can it act?
What happens when your agents talk to other companies' agents. Brewline's oat milk runs low, and its agent asks two suppliers' agents for quotes. We look at the Agent2Agent (A2A) standard, how it differs from MCP, and real examples from Tyson Foods and Gordon Food Service, Microsoft, SAP, ServiceNow, Amazon and Google's Agent Payments Protocol. Then the new risks, and the decisions to make before your agents talk to outside agents.
Skills give an agent a shelf of training manuals: small folders with instructions, templates and scripts that it opens only when a task needs them. We look at what a skill is, how it differs from tools, prompts and retraining, real examples (built-in document skills, Rakuten, Box, Notion, Canva), a Brewline skill for the monthly cafe report, and the risks and decisions that come with installing skills.
Explore how AI agents use memory: short term, episodic, and long term—guided by vector databases and retrieval augmented memory to recall, learn, and personalize interactions while addressing privacy.
The Brewline Care Agent gets a memory. With real agent runs and a real embedding search, you see what short-term memory needs to keep from a long chat (pinned facts), and how episodic memory turns a vague complaint into a clear customer story. Download the Brewline Memory Experiments workbook to see every run and score.
Two memory problems, run for real: an old FAQ that still wins the search (stale memory), and a kind barista note that should never have been stored (private memory). You see how dates, owners and retiring old versions fix the first, and why the second needs a decision before anything is saved. Download the Memory Policy template to make the five decisions for your own agent.
Explore planning and reasoning in ai agents, from chain of thought and multi-step planning to task decomposition and planning graphs, to enable coherent, self-correcting decision making.
Brewline's coffee supplier raises its price by 18 percent, and the renewal deadline is close. We give a real agent the same planning job twice: first we ask for an answer straight away, then we ask it to plan first. You see a sensible one-shot answer with no data behind it, and a professional task graph of 11 subtasks with a hidden flaw. Part 2 lets the agent plan and act.
Now the agent plans and acts, with access to Brewline's systems. Every one of its 51 steps is logged. Twice, real facts force the plan to change: data from two cafes is missing, and an office contract clause hides an earlier deadline. You see how the agent replans, what its decision brief recommends, and four signs of a plan a leader can trust. Download the workbook with the task graph, the full log and the brief.
Explore recursive and iterative prompting that lets AI agents reflect, self-correct, and refine outputs through multi-pass reasoning and prompt chaining for high-stakes tasks.
A real review chain for Brewline: one agent writes a subscriber email, a separate reviewer scores it against a five-point rubric, and the writer revises. You see every score (15, then 22, then 23 out of 25), five subject lines ranked by the reviewer, a real disagreement between the agents, and why every loop needs a stop rule. Download the Review Loop rubric template and the workbook with all drafts and scores.
Explore learning mechanisms that let advanced AI agents evolve across sessions and tasks. Learn few-shot and in-context learning, feedback loops, and reward-based adaptation to improve over time.
A real learning run for Brewline's Care Agent. In round one, Mira, the support lead, corrects five of eight replies. The agent turns her corrections into five written lessons, one of them its own idea. In round two, eight new emails with traps go to two agents, one with the lessons and one without, scored blind: 7 ready to send, 1 to fix and 0 wrong with the lessons, versus 4, 3 and 1 without. You also see a rule nobody gave the agent, and why a leader reviews the lessons file. Download the workbook with every email, reply and score, and the one-page Learned Rules Review Checklist.
Learn how ai agents coordinate intentions, negotiate goals, and collaborate with humans and other agents to achieve shared goals through turn taking and message passing.
Explore how ai agents act autonomously—sending emails, updating calendars, or placing orders via APIs—while safety layers, human-in-the-loop checks, and permissions ensure responsible outcomes.
Explore how embodied agents translate digital intelligence into physical and virtual actions. Learn examples from surgical robots, warehouse automation, avatars in VR/AR, and metaverse companions, plus safety and trust challenges.
Identify modular components of modern AI agents and compare frameworks like Autogen, crew I, and Land Graph to design scalable, reliable agent architectures.
Learn to build a simple AI agent from goal definition to testing, integrating tools and memory, and choosing architectures to enable reliable, proactive agents.
Design and implement a customer service escalation agent by mapping intent to architecture, using Cry or Lang graph to analyze tone, check history, and generate replies.
Explore modular ai agents for leadership, from web-searching research assistants that summarize content to meeting assistants that extract decisions and emails, and reflective coaching agents that support leadership reflection.
Hands-on: build it in Claude. A real run from our Claude for Business Leaders course. Will runs a bike shop. He gives Claude 40 customer emails and the shop's policy sheet and asks for a customer FAQ page. Watch how Claude flags the questions the policy does not cover instead of guessing, just like the escalation agent in the earlier lecture. The slides were first made for that course, so the footer still names it. Download the two-page FAQ kit below and try it with your own emails and policies.
Hands-on: build it in Claude. A real run from our Claude Skills for Business Leaders course. Camille leads customer support at an online running-shoe shop. She teaches Claude her way of working with a Skill: her voice, her reply structure, her return policy, and a rule to write NEEDS HUMAN when the policy is silent. Then the policy changes and we test the skill again. The opening slide still says Use case 3 because it comes from that course.
Master how multiagent systems achieve scalability and modularity by coordinating specialized agents through collaboration, delegation, and structured communication, supported by diverse frameworks and no-code tools.
Explore active research in artificial intelligence agents across planning, memory systems, reward modeling, robustness, and safety layers like planning guards to enable dynamic replanning.
Explore trends in personal and decentralized AI agents, including persistent memory and open-ended autonomy, and learn how leaders design modular, ethically governed systems.
Well done, you finished AI Agents for Leaders. Here is your reward: practice deeper skills with AI, with these five tools from AI for Happiness. The links are also under Resources.
Try AI for Happiness for 1 month, for free:
https://aiforhappiness.com/invite/home?ref=udemy-course
A small favour: if this course helped you, please leave a short, honest rating and review on the course page. Say what you learned and what you used. It helps other leaders find the course, and it tells me what to improve.
Follow me on LinkedIn:
https://www.linkedin.com/in/anamariaolteteanu/
AI agents are transforming work—and leaders need to understand how to harness them.
This course is designed for professionals, innovators, and decision-makers who want to lead confidently in the age of AI agents. You’ll gain a clear, practical understanding of what agents are, how they work, and how they’re being applied across sectors like healthcare, enterprise, and education—without needing to write code.
In this course we will explore:
What makes AI agents different from bots or automation scripts
Core capabilities: memory, planning, tool use, communication, and learning
Use cases: from care coordination and workflow agents to productivity and coaching companions
How to evaluate frameworks like LangChain, CrewAI, and LangGraph (without needing to build them yourself)
Leadership insights on agent architecture, team readiness, safety, and ethical design
Future trends: multi-agent systems, embodied agents, and strategic AI roadmaps
You’ll leave with the ability to speak the language of AI agents, evaluate tools, identify valuable use cases, and guide technical or business teams in responsible deployment.
Who This Course Is For:
Innovation and digital transformation leaders
Product and strategy professionals exploring AI capabilities
Business professionals who want to lead AI adoption — not just watch it happen
Consultants and team leads who need to bridge business and AI development teams
AI professionals, data scientists and programmers wanting to understand the principles of AI agents
No coding skills required. Just a sharp mind, a curiosity for emerging technologies, and a desire to shape the future of work.