AI Capability Framework: Strategic Delegation, Direction & Agentic AI for Leaders

1. Introduction to Agentic AI Model: Let us understand different AI models or a genetic frameworks. We have LLM, which is by itself really smart. You can ask question, it can answer. But just like the way intern can't open files, can't look up the latest rules, can't remember what you told them yesterday. That's limitation. An augmented LLM is like upgrading that intern into a knowledgeable worker or into a knowledge worker with three extra superpowers, retrieval tools and memory. Let's break down the powers retrieval, access to external knowledge, like giving intern access to your company's file cabinet and Google search. This is a similar capability that the augmented LLM retrieval part has. Let's think about a healthcare example. Prior authorization agents pull the latest payer rules before filing a request. Mortgage example loan setup agents retrieve updated interest rate rules before drafting the documents. Lent, remove waste of manual lookups, reduce rework when outdated rules are used. The second is tools, ability to act inside systems. Instead of just reading policies, the intern can now log into portals and execute task. Agent logs onto a payer portal, submits the pre automatically. Agents call the Credit Bureau API to fetch credit scores. So that is augmented LLM. We want to reduce human handoff so that the cycle time becomes faster. The third important power is memory, ability to remember context over time. The intern keeps a notebook to keep everything from past meetings and doesn't ask you for the same question twice. Let's take an example from healthcare. Patient Experience board remembers your last billing query and continues smoothly. Loan assistant recalls that the borrower's income proof was missing from the last meeting. Improve first pass yield. The objective is to reduce errors due to missing context. When you combine retrieval plus tools plus memory, you don't just get a chatbot. You have an agent that can look up, act, and adopt fewer handoffs, less waiting, and less rework, or leading to faster cycle time and higher accuracy. Today, a nurse handling denial spends 20 minutes. She looks up the payer rules, retrieves it, logs into three system tools, and remembers the past denial cases from the memory. Now, imagine an augmented LLM doing that in seconds. That's the leap from AI as a helper. AI is a team member. A plane LLM can take inputs and give you the output. That's useful, but it's limited. It forgets the context. You can't always access the latest data, and it isn't designed to use external tools. Whereas an augmented LLM fixes that by connecting the model to three key extensions, retrieval tools, and memory. Retrieval is about access to knowledge. LLM can query external database, knowledge systems, or documents to fetch facts in real time. Healthcare, pulling up the patient history before drafting the summary. Mortgage retrieving the updated credit scores and policies before res scoring. Tools have the ability to act. LLM can call APIs, trigger workflows and run calculators. Invoke a fraud detection tool, trigger an RPA board to reset the password. That is the capability. Memory is continuity over time. The LLM can remember prior interactions, preferences, and past actions. Remembering brand tones across multiple press releases. Recall a student's learning progress to tailor the lessons. Together, these augmentations turn the LLM business grade collaborator, one that can recall history, pull facts, and take actions, not just generate text. Think of a plain LLM as a bright intern. It can answer questions but forgets everything after the meeting. An augmented LM is like giving the intern access tools and memory that can ensure that it's becoming more valuable and enterprise ready. 2. Leaders Thinking with AI: If AI is a co pilot in every tool, then we need to train the pilots, not the passengers. Hence, we are all here. The four Ds of AI Capability Framework. This framework is designed to answer critical questions for leader. How do we build AI fluency at scale, not just AI usage. The four Ds provide a leadership oriented capability map. The first one is delegation. Not everything must be handed over to AI. Leaders must define what remains inherently human judgments, values, relationships, versus what can be codified and scaled with AI. This decision is strategic, not operational. Misplaced delegation leads to risk. Thoughtful delegation creates leverage. The second D is direction. AI is only as effective as the clarity of the problem it's asked to solve. Many failed AI pilots stem not from weak algorithms, but from the poor articulation of the business needs. For example, in the finance services, if the ask is framed as reduce fraud, AI struggles. But if it is framed as detect anomalous transactions, above $10,000 within 5 minutes, the system thrives. Eaders must insist on precision in framing the question. The third D is detection. AI will give you answers, but leaders must build the muscle to interrogate them. Not everything that is possible is correct. This is where the senior teams bring a critical evaluation. Cross checking AI's output against the domain knowledge, compliance, and business outcomes. Finally, diligence. Deploying AI without responsibility is dangerous. Ethical guard rates, data governance, and regulatory foresights are not back office items. They are the boardroom priorities. For transformation leaders, diligence is the line between scaling responsibily and creating systematic risk. This framework is less about how to use a tool. It is more about how to lead it with responsibility in an AI driven enterprise. Now that we have explored the four Ds, let's connect them to our organization design. How leaders can embed these capabilities into teams, governance structure, and culture. We know how to use AI, but more importantly, you will know how to think with AI. This statement is at the heart of why we are here today. Most organizations are teaching people how to use AI, how to prompt, how to run a model, how to automate a task. That's useful, but it's not transformative. For leaders, the real differentiator is learning to think with AI. That means shifting from seeing AI as a tool in the corner to making it a thought partner in how problems are framed, decisions are made and opportunities are spotted. Consider this. Using AI is like asking E to create a summary of a market report. Tinking with AI is asking, what pattern does it reveal about customer behavior and what business models would emerge from it? The first is tactical. The second one is strategic. In transformation programs, this distinction separates a short term efficiency play from an enterprise wide reinvention. L eaders who only know how to use AI will get productivity gains. Leaders who know how to think with AI will reimagine the industry. So as you reflect on your role, ask yourself, am I treating AI as a calculator or as a co strategist. That mindset is what will shape competitive advantage in the next decade. Let's now look at how this mindset translates into capabilities. The specific fluency leaders must build in their teams to move from using AI to thinking with AI. 3. What is Delegation: Dedication is the first and the most crucial competency in AI Capability Framework. For transformation leaders, the real question isn't whether AI can do something. It's whether it should do it. Let's bring this to life with two domains where many of you have a direct oversight, customer service department and claim processing. In customer service, AI can handle initial triage, routing queries, recognizing VIP status, pulling up customer histories. That works best delegated. They are repetitive, rule based, and prone to human error if done manually. But when it comes to handling emotional escalations, retaining a high value client or making a goodwill exception, that's human territory. Delegating that to AI would undermine trust and relationship by our human clients. In claims processing, AI excels in validating forms, checking policy coverage, and flagging anomalies for fraud. Delegating these steps reduces the cycle time dramatically. But final settlement discussions, dispute resolution, and policy exceptions still require human oversight because they blend judgment, empathy, and risk. The competency of delegation is about drawing the line. Where does AI give us speed, accuracy, and scale, where does human intervention preserve trust, nuances, and responsibility? Leaders who get this wrong either waste human capacity on low value work or expose the enterprise to reputational and compliance risk. Let me clarify what AI Capability Framework is all about and what it isn't. This is not about memorizing the top ten prompts or chasing the latest hack on Chat GPT or copilot. Those tricks became outdated in weeks. What last is the capability, the habit, the skills, the judgment you bring to every AI interaction. Think of it in the way that you think about N or six Sigma. Tools evolve, but the mindset and the discipline remains. 4. Delegation: Problem Awareness: Delegation and problem awareness. Before we talk about AI, let's talk about us. The cornerstone of a good delegation isn't technology. It is clarity. Success starts by knowing exactly what we are trying to achieve. Too often, teams rush to apply AI without defining the real goal. The result faster outputs that don't actually solve the business problem. So before involving AI, pause and ask four simple but powerful questions. What exactly am I trying to accomplish? Do I want to reduce the claim cycle time, improve the first call resolution in my customer service department? What exactly am I looking at? What does success look like? It is shorter turnaround time, fewer errors, higher NPS, or lower cost to serve. We need the clarity for it. What kind of work is required? Is it simple but time consuming, like summarizing a 20 page policy? Is it uncertain and exploratory, like finding a new fraud patterns, or is it judgment intensive, like handling an unhappy VIP customers? Where is AI best place? Where must human stay in charge? AI can draft and summarize and classify. Human can decide, interpret, and take accountability. Delegation is not about offloading work. It is about breaking down complex workflows into parts and assigning each part to the right partner. Human or AI. When we start with clear goals, delegation becomes strategic. That's where efficiency and effectiveness comes together. Thank you. I will see you in the next lesson. 5. Delegation : Platform Awareness: Delegation and platform awareness. Now, once we are clear about our problem, the next capability is platform awareness. Not all AI platforms are built in the same way. Some are lightning fast but shallow, some are slow, but better at reasoning. Others are tuned for creativity, crafting marketing copies and generating image, and some are optimized for data accuracy, better at handling structured content like invoices or claims. This matter because choosing the wrong platform is like giving a hammer to a searcher. The tool itself isn't bad, but it's the wrong fit for the job. For example, if you need speed and scale, like scanning thousands of customer emails for sentiments, you might choose a lightweight classification model. If you need judgment and reasoning like analyzing a fraud claim, you need a reasoning model that can explain its logic. If you are in a creative problem solving, like rethinking your customer's self service flow, you may want a generative model that can brainstorm with you. And here's the important part. Don't lock your team into just one platform. The field is moving too fast. Encourage experimentation. Give your analyst and Ops managers the chance to try two or three platforms side by side. Transformation leaders learn this lesson with RPA, BPM, and lean tools. The strength isn't in a single platform. It is about knowing which platform fits the problem. So here's the mindset shift. Delegation isn't just what task do I give to AI? It is also about which platform is best suited for this task. That's how you combine problem awareness with platform awareness to make AI a strategic partner instead of a blank box. Thank you. I will see you in the next lesson. 6. Delegation : Task Distribution: Delegation and task distribution. Once you are clear about your goal and the AI platforms available, the real art begins deciding how to distribute the task between human and AI. This isn't about replacing people. It's about balance. Think of three buckets, automation, task AI can handle safely and repeatedly. These are the routine rule based steps where scale matters more than judgment. Auto extracting claim numbers from the PDF form, tagging support tickets by category before routing it. Augmentation is about task where the human and AI work side by side. Here, AI helps to speed things up and expand the option, but the human stay in control. A policy analyst asking AI to draft variation in the clause, then refining the one that fits. A customer service leader co developing email templates with AI, then tailoring the tone for sensitive cases. Human only judgment, the task that should never be delegated. These are the decisions that demand human awareness, context, accountability, and nuances. Deciding whether to deny a borderline insurance claim. Or making a compliance interpretation that carries regulatory risk. And then there's a fourth category worth highlighting, low value repetitive work, the things no experts should spend their time on. These are perfect candidates for AI agents. Things like generating meeting notes, preparing the first draft of reports or consolidating routine metrics. So the guiding question for you as transformation leaders are, which part of your process are ready for safe automation? Where can augmentation create more value by enhancing human performance? Which area must remain human led to protect judgment and trust? And what repetitive work can we confidently hand off to AI agents? Getting this balance right is what separates organization that simply use AI from those who truly create value with AI. Thank you. I will see you in the next lesson. 7. Second D: Direction: The second D of the AI capability framework is direction. Let's dig a little deeper into this competency of direction. Think about how you can brief a new team member. You don't just say fix the SLA issue. You provide the context what the SLA is, why it matters, what tools are available, what exceptions to watch out for, and what the final deliverable should look like. That's direction. Now, replace that team member with an AI system. The principle doesn't change. If you skip context and clarity, the AI drives vague or incorrect results and the blame shifts unfairly to technology. When in reality, it was a communication failure. Direction is about how you communicate with AI. It sits at the heart of almost every human AI interaction. Clear direction transforms AI from a black box into a partner. Poor direction reduces it to a noise. Let's make this real with an example across domains. If you're thinking about a healthcare provider service, instead of asking AI, summarize the patient node, direct it in a way, it's telling that summarize the last three visits, highlight the medication changes and upcoming test. Let's take an example from claim settlement process. Don't say review this claim. Instead, check this motor claim for fraud risk by comparing repair cost against historical patterns, and flag anomaly is over 20%. Let's take an example from mortgage. Don't say process of application, but say extract the income and employment details, flag the missing documents, estimate the approval probability based on the underwriting rules. If I think about media or communication department, instead of saying draft a press release, you can be more specific in your direction by saying draft a 400 word release for a B to B audience with a confident tone, highlighting regulatory and compliance benefits. The competency of direction ensures AI delivers work that is effective, efficient, ethical, and safe. The very outcome transformation leaders care about when you master direction. AI stops being a guesser and it becomes a true contributor. 8. Direction: On isn't about clever wording or tricky prompts. Think of it like writing an SOP or a process document. You are translating the business logic into AI readable instruction. Just as when you onboard a new analyst, you don't just say figure it out. You tell them what the outcome should look like, what approach they should use, and what tone should be maintained. Poor direction results in rework. If AI delivers a wrong tone, wrong format, or missed logic, someone has to fix it. That kills adoption. Every bad output chips away the user's trust. Consistency comes only when the direction is strong. Direction is also about governance. It's how you keep AI aligned with business rules without writing new code. In every transformation program that you have led, you have already done this. You have defined what success looks like. You decide the approach or the method, and you align the tone and the behavior. So these becomes your three important pillars of direction. So now, simply applying the same rigor is when you are interacting with AI. Let's start with the first dimension of direction. That is product direction. Think about how often AI disappoints you, not because the model is weak, but because we didn't tell it exactly what we wanted. AI is not a mind reader. If you leave it guessing, the output is often missed the mark. Product direction means answering four simple yet powerful questions up front. What's the context for this work? Exactly, what should the AI do? What format should the output take? Who is the audience and what style is appropriate? Let's make this real with some HR examples across some of the subprocesses. Talent acquisition team. Instead of giving a direction, telling the review the resume, give a clear product direction. Summarize this resume in three bullet points. The section should include relevant skills, relevant experience, and evidence based strengths. Avoid inferring personalities or demographic traits. Here, the AI now knows what to include and what to avoid. Let's take an example from performance management, instead of saying, write feedback for this employee, give a product direction, draft behavior based performance feedback. Structure it into an achievement and strengths, development areas and next steps. Use a neutral supportive tone which is suitable for a mid review. There is no guessing the AI has a blueprint now. Let's think about the policy communication. Instead of asking the AI to explain the new leave policy, give product direction. Write 150 word employee communication, explaining the new leave policy. Use simple language, avoid HR jargons, and end with two clear action steps. The audience is all the employees of our organization. You can give the name of the organization if you want. This ensures clarity, tone, and structure that matches the HR communication standards. We will take one more example from employee relationship team. Instead of asking the AI, summarize this complaint, give a clear product direction. Summarize the employee's complaint into factual chronological format, include dates, action taken and involved parties. Please do not interpret emotions or assign blames. This keeps the output compliant and investigation ready. So why is product direction important? It is because when you provide a clear product direction, AI has a blueprint. You are not leaving it to guess. You are setting explicit requirement so that the output aligns with your goal and your standards and your audience, once you are clear on the product direction, the what, the next step is to guide the how. That's where the process direction comes in. I'll cover that in my next class. 9. Process Direction: Now let's look at the second dimension of the direction. Process direction, the how. In complex or regulated environments, methods matter as much as the outcome. Think about your own team. Sometimes you don't just care about the job getting done. You also care about how it is getting done. The same applies in EI. With process direction, you are guiding AI's approach. The results are not just fast but reliable and compliant. There are several ways to do this. General guidance, like handling someone's manual, step by step instruction like giving a recipe, worked examples showing, here's how I do it. This matters because AI already has a broad training, but it doesn't know your context unless you explain it. So you want to answer questions like what data source should it draw on? Which issues must be addressed and in what order? What workflow or analysis style should be used? Let's ground this with example from your domains. Healthcare providers services. Instead of summarizing patient notes, say, summarize the last three visits in a chronological order, highlight the medication changes, and then flag any upcoming test. When you give process direction, you are not micromanaging. You are shaping the AIs method to reflect your business rules. That's how you prevent errors, reduce rework, and ensure compliance. Thank you. I will see you in the next lesson. A 10. Performance Direction: Performance Direction. If there is one takeaway from this module, it is this. AI is not a database. It is not a vending machine. It does not simply store facts or spit out a fixed answer. AI is an interactive system, or just like people, its behavior changes based on how you guide it. That's where performance direction comes in. Performance Direction is about shaping how you want the AI to think, respond and present itself. It's not about what AI produces. That's product direction. It's not about how AI should execute the steps. That is process direction. Performance Direction is about the personality and the behavior of the output. Before you start working with AI, ask yourself four questions. Do I need an assistant that narrows towards one correct answer or a partner that explores multiple possibilities? Do I want the AI to challenge assumptions or simply follow my instruction precisely? Should the output be detailed and rich or concise and to the point? Do I want the reasoning step by step or just the final polished answers? These choices dramatically influence the quality and the usefulness of the results. Let's make this real within HR. Performance Direction, let's look at some examples. Instead of saying draft a response to an employee complaint, the performance direction gives you options like write a calm, neutral policy aligned response, which is suitable for formal communication. Or you might give a direction telling, write a supportive, empathetic, two sentences acknowledgment before HR investigation begins. Same task, completely different performance expectation. Let's take one more example from the talent acquisition team. Instead of saying AI, write interview feedback, you set the direction, write a structured feedback in competency based format. Avoid personality judgment. Or you can say write a short manager ready summary highlighting strengths, risk, and hiring recommendation. Performance Direction ensures fairness and clarity in hiring communication. Let's take an example from the LND department. Instead of saying explain this policy, you can specify the behavior. Explain this policy in simple and learner friendly language as if teaching the new hires, or explain this policy in a detailed manager level language with examples and implications. This changes the tone, the depth, and the level of complexity in the output that comes out. Performance management, instead of saying, summarize this appraisal, you can guide how AI should behave. Summarize the appraisal using action oriented language and avoid generic adjectives. Summarize with coaching focus tones, prioritizing growth amendation. Performance Direction shapes whether AI sounds like a coach or a policy administrator. Why is this an important skill that you need to know? Because without it, AI behaves in a generic way. With it, AI becomes a precision tool that adapts to your audience, whether they are employees, managers, HR, leadership, or the new hires. It understands the purpose of communication. Is it a feedback, communication, coaching or documentation? It understands the tone, should it be empathetic, firm, neutral, or formal? The output can be defined in a brief, structured R detailed style. So when you combine product direction, what you want, process direction, how to do it, and performance direction, how it should behave, AI is just not an assistant. It becomes a thinking partner that produces HR outputs that match your exact standards. That's when AI begins to deliver true business value. Now that you understand how to guide AI with direction, the next step would be detection, developing the discipline of evaluating AI outputs critically, so you never accept an answer at its face value. We will cover that in the next video. 11. Preparing for Process Direction: Now let's look at the second dimension of direction, Process Direction, the how. In HR, how work gets done is just as important as the final output. Think about your own HR team. It's not enough that the task is completed. It must be done accurately, fairly and in compliance with the policy. The same applies when working with EI. Process Direction guides the AI methods so that the results are not just fast, they are reliable, structured, and safe. There are several ways to give process direction. General guidance is like handing over someone a playbook. Step by step instruction is like giving a procedure or an SOP manual. You can also give some worked examples like showing how the HR usually does this. This process of process direction is very important because AI has broad training, but it doesn't know your HR context unless you tell it. So with process direction, you clarify what data should it use first? What must be checked in sequence? What analysis style is appropriate for HR? Which steps are mandatory for compliance? Let's ground this with examples from HR subprocesses. For a talent acquisition team, instead of just saying screen this resume, you have to give a process direction. First, extract the job relevant skills, then map them to the job requirement, then identify evidence based trends, and finally list missing or unclear information for the recruiter to follow up. As you can see, this shapes how AI evaluates candidate fairly and systematically. If I have to take about complaints and investigation from employee relationship process, instead of saying summarize this complaint, we give a process direction. List the events in a chronological order, highlight documented facts only, then identify areas that require additional clarification. Do not interpret emotions or assign blames. This prevents inaccurate assumptions. The next example from performance management team could be that instead of draft performance feedback, give process direction. Start by identifying measurable achievements, then link behavior to competency, then outline development areas using neutral language and end with the next steps aligned with the performance framework. You can see that we have given very detailed instruction, the process that it has to follow. This enforces fairness and consistency. Moving to the policy communication, instead of saying explain this policy, if we are giving a process direction, we will say, break it down into three parts. What the policy means in simple terms. When does this policy apply? Where it doesn't work? What employees need to do next, avoid any legal jargon unless necessary. This ensures clarity and accessibility. Create a training summary. Instead of this, I can give a clear process direction, extract the key learning objectives, then summarize employee feedback into themes, then flag any recurring skill gaps that need follow up. This improves insight generation. So you might have understood why process direction is important. When you give process direction, you are not micromanaging AI. You are shaping the method it follows to match the HR standards, policies, and compliance expectation. This is how you prevent errors, reduce rework, and maintain fairness. This also helps in improving consistency, ensure policy alignment. Process Direction is what turns AI from a fast tool into a reliable HR partner. I will see you in the next dimension. 12. Weak Direction: A weak direction would be like, write a customer email about delay. The AI output would be like, dear customer, your product is delayed. We apologize. A strong direction would cover product process and performance. So you will say, write a three line email to a VIP customer who has experienced a three day delay in the delivery. Mention the reason that as vendor issue and offer a 10% discount coupon and keep the tone warm but professional. The AI output will be very different. It will say, Dear Mr. Ramesh, we are sorry for a three day delay caused by a vendor issue. As a token of apology, here's a 10% discount for your next order and thank you for your patience. Same AI model, same technology. The only difference was the direction. And hence, it's an important skill to build. If someone says it's just prompting or it's just prompt engineering, the answer is no. Direction is about operational clarity. Just like RPA fail when we automate with wrong steps, AI fails without clear direction. This discipline prevents misuse and it helps in building the trust. In your last few transformation projects, how often did you miscommunicate creating rework, delay, and cost overrun? AI is no different. The cost of poor direction is same. Rework, frustration and lost confidence. The payoff of a good direction is same too, faster cycle time, higher quality and reliable adoption. 13. Third D: Detection: Now that we have covered direction, let's move to the counterpart detection. If direction is about clearly communicating what you want, detection is about quality control, evaluating whether what the AI produced is actually fit for HRuse this is one of the most critical competency as leaders, because no matter how advanced the model is, AI can and will make reasoning errors, misinterpret your context, produce biased or non compliant freezing. It can overlook important evidence or generate responses you didn't expect. Detection requires you to pause and ask three questions every time. Is this output valuable or problematic? Does it show the strength of AI or expose its limitation? Is it ready to use or does it need refinement before it reaches employees or managers? Doing detection well requires two things. Your HR domain expertise, your ability to judge quality in HR context. So let's take an example. You can instantly recognize if AI has misinterpreted a skill or inferred something discriminatory. You can see whether the feedback is behavior based or personality based. You can tell if the complaint summary is factual, neutral, and investigation ready. For compensation and benefits department, you know if AI misunderstood eligibility criteria and miscalculated salary components. We need to understand the limitations of EI, knowing where AI typically falls short. EI sometimes fabricates details that were never in the document. You may write feedback that sounds confident but violates HR policy or fairness standard. AI can misinterpret tones becoming too harsh or too casual. AI may overlook handwritten notes and H cases or exceptions. Let's look at an example. AI drafts an announcement about a new leave policy, but the wording unintentionally sounds provocative. Detection tells you that the tone will damage the trust if published. 14. Agentic AI Activity: Let us understand how you can have your personal AI fluency plan. I'm going to give you some activities which can help you assess your current skills and how you can build your competency around the FOD AI capability framework. So the step one is to assess your current skills with examples. So let's say you want to rate yourself on the FOD competency. You might say that I'm a novice person doesn't have much idea about it. You might say I'm still developing skills to get there, or I'm a confident user of the framework. So the template is very simple. You have the competency in the first column. You rate yourself. It's a self assessment. What are the strengths you feel that you have in this space, and what is the gap that you want to fulfill? So for example, you are a nos user. What do you feel are the gaps and you want to fill it up? So let's take some examples that I have filled up. So one of my friend was working in the HR domain for the talent acquisition team. So we rated down all the four competency in the first column that is delegation, description, detection and diligence. So for the first competency of delegated her as developing, it is a self assessment that she had rated. I'm just helping you understand how the table has to be filled up. So delegation, which is about developing, the strength that she has is that she can tell AI to summarize the SVS. And the gaps that she fell is that she's not good at deciding when AI should not screen a candidate. So as you can see on the screen, this is the template that has been filled up by her, right? So it becomes easy for you to understand that for each of the Ds in the competency framework, you are going to do yourself rating, identify those strengths, and fill the gaps. The second one is description. So she has marked herself as developing. She's good with writing some basic prompts, but she struggles to give examples and she's having difficulty in getting the correct tone and constraints. Detection is the third D of the AI capability framework. She rates herself as nos and can spot obvious mistakes, but it's hard to identify the bias in the AI generated candidate summaries. Diligence. She marked herself as developing score. She's aware of the data privacy rules, but not sure how to decide what candidate data is safe to upload. Now let's take one more example from the loan underwriting team. So again, the competency, rating, strength, and the gaps. So this is the other friend of mine who has rated it. So Diligence marked it as developing good at asking AI to extract key financial data, but unsure when to let AI recommend versus only summarize. Description, confident in managing the description part because can give clear instruction on tone and format. Need to work on multi step prompting. The third stage is detection. It is at a developing stage, good and can check numbers, but need to detect hallucination of financial ratios. The loan underwriting officer for the Diligence part mentioned or scored herself as novas, aware of compliance risk, but not confident about the data governance requirement. So let's go to the next example that we have about the healthcare provider service. So again, force Ds are listed, delegation description, detection and Diligence, and they have scored for each of the competency. They have rated down the strengths for each of the competency and the gaps that need to be filled up. So as you can see, it's important for us to fill up these competency self assessment scoring. This can help us in understanding what we need to learn more. The step two would be reflecting on the three modes of AI interaction. Do you remember the three modes? Yes, it's automation, augmentation and agency. You have to mention what is your comfort level. So the template is again, very simple, as you can see on the screen, that is the modes are listed, automation, augmentation and agency. What is the comfort level? It is high, medium or low, and your commentary. So let's understand with help of an example. So from HR in an LND team, the person has rated that automation, the comfort is very high, uses AI to generate MCQ questions and summaries. Augmentation, the comfort level is medium. They have tried to use it for co creating learning journeys. Agency, it is low, haven't used AI agents for auto generated reports, so hence not comfortable with this part of mode of interaction with AI. Now let's move to the next example of identifying priority areas, right? So again, here, competency, I can prioritize first. First, assessing my skills in the four D competency, that is delegation description, detection and diligence, I will focus on description and detection. So you are setting a priority wherever you feel you want to first learn this competency in detail. These two competencies will have a highest impact on my professional performance because I frequently work with AI for writing summaries, drafting HR content, analyzing documents, and supporting the decisions. The second question is why these competencies matter? I need a detailed explanation. So here, why does description matter? Why is that as a priority of learning these competencies first? I have noticed that when my prompts lack detail, hit outputs that are too generic, mismatched in tone, incomplete or misaligned with what I want. Clear instruction, high quality output, but vague instructions give me random inconsistent results. Description is my biggest bott neck. How I will benefit, HR processes will become more consistent. Example, interview guides, JD creation. Banking style summarization tasks will become more accurate. Example, policy summarization. A healthcare provider service agent might say that the healthcare administrative tasks like as summaries will avoid jargons and errors. If the example from media content will be finally match the tone requirement of internal communication and newsletter. What I must stop doing? I should stop writing one line prompts, not giving examples, not defining tone or audience, not specifying what not to include, expecting AI to magically understand the context. So these are the things I should stop doing. Now, mooing, why did I take the priority of detection as my second priority that I want to work with. AI often produces fabricated details, incorrect reasoning, over confident claims, and biased language. Detection helps me evaluate output critically instead of believing everything AI sees. How I will benefit in HR, I will prevent bias to resume summaries. In banking, I will verify numbers instead of trusting the hallucinated data. For a healthcare provider service, I will benefit by saying that I will avoid accidental clinical interpretation. From a media point of view, you might see that I will prevent AI from generating unverified claims or stats. What I must stop doing accepting the first draft as final, assuming numbers are accurate, forgetting to ask AI for the confidence score. Letting AI infer demographic details like A, gender, et cetera, not cross checking against source document. The third step would be the most valuable skill to build within each competency. You have to be deep, specific and domain relevant. So priority number one was description. So this is the skill to build. This is the first skill, giving multi layered instruction. That is one skill that I want to build in description. Let's take an example. Instead of saying summarize this resume, I would write summarize this resume, focusing only on the skills and the experience relevant to the job. Do not infer age, gender, personality, or educational qualities. Provide three strengths, one development areas and justify everything with evidence. Specified tone and audience. Media communication. The example is write 200 words investor update in confident data back tone. Avoid objectives like amazing and revolutionary. Use short sentences. The skill three is that I want to be context rich prompting. Example, could we summarize this patient case for administrative documentation only. Do not diagnose, recommend treatments or interpret systems. Giving examples and constraints, create assessment summary that follows each format or this format and excludes assumption. I want to focus my second priority on the description skill set. So the skill one that I want to build is giving multi layered instruction. A good HR prompt tells AI what to focus on, what to avoid, and what structures to follow. So instead of saying summarize this resume, a strong HR prompt is summarize this resume, focusing only on job relevant skills and experience, provide three strengths with evidence from the resume. One development area, also evidence based rules do not infer a, gender, personality, or attitude. This avoids bias and gives AI a clear direction. The skill tool is to specify tone and audience. Different HR tasks require different tones, managerial, empathetic, neutral, or corrective. So the example here could be rewrite this policy update in a clear, simple tone for employees. Avoid technical HR jargons. Use short sentences and give a message supportive. Tone direction prevents miscommunication. The skill three I want to build is about context rich prompting. AI becomes far more accurate when you provide HR specific context. Summarize this employee complaint in HR investigation, include only factual events, dates, and actions. Do not assign blame, interpret emotions or offer solutions. Adding context prevents AI from making unsafe assumptions. Skill four is about giving examples and constraints. Showing AI what good looks like helps it to match your expectation. Example, create a performance review using the same structure as this example, key achievements, strengths, development areas, such as next task, and keep the tone neutral and avoid unsupported claims. This example and constraint keeps the output consistent and unbiased. Now, detection, right? Again, I'm taking some HR only examples. So if detection is the skill that you want to build, detection in HR means evaluating AI output carefully, critically and safely. So you never rely on assumption, bias or unsupported statements. The scale one would be about asking for reasoning and verifying steps. AI sometimes reaches conclusion that looks logical but are not grounded in the source material. So when you're screening a resume, instead of accepting the output blindly, request transparency. You prompt should be show exactly what resume lines, support each strength you have identified. If something is not directly supported by the resume, please highlight it. Why does this matter? Because it prevents AI from inventing skills or inferring personality traits. The skill too could be about identifying bias and harmful assumptions. Bias can only easily appear in HR outputs if not monitored. Let's think about a performance review writing. If AI writes, John seems unmotivated. Ask the AI, what evidence from the document behavior supports this? None exist, instruct, rewrite only behavior based factual observation. Remove all personality interpretation. It ensures feedback remains fair, defensible, and compliant. Checking domain accuracy. AI can misunderstand HR policies, frameworks, and legal boundaries. If I think about a policy interpretation, ask rate your confidence in each HR term you need, example, constructive feedback, misconduct or PIP. Highlight anything below 80% confidence. Then manually verify everything or anything that you feel is uncertain. Incorrect HR terminologies can create legal and ethical risk. Contrasting outputs for tone, clarity, and appropriateness. Sometimes you need to identify different styles to choose. Example, from employee communication would be ask EI to generate three versions. One conservative, which is using a formal minimal emotions. Second, a term supportive version, empathetic and bomb tone, and the third one could be a neutral, that is a concise version, straightforward facts. Then evaluate which fits the situation. For example, announcing a new policy versus addressing a grievance. Different HR scenarios require different tones, and contrasting version helps you choose the safest and the most effective tone. Which AI interaction mode I will prioritize. The primary mode to improve is augmentation. 15. Thank you for Choosing AI Agents class: As we come to the end of this class, I want to pause for a moment and say something simple and sincere. Thank you. Thank you for being my students. And thank you so much for taking out time and completing this course with me. You didn't just watch the lesson. You engaged with complex ideas from foundational language and models to fully autonomous agent system. You explore what makes an agent different from the workflow. You learned how augmented LLMs combine retrieval, tools, and memory and reasoning. You examine real architecture patterns like prompt chaining, routing, optimizer, parallelization, and orchestration. Most importantly, you now understand something powerful. Agentic AI is not about adding complexity. It's about adding the right level of autonomy at the right time for the right problem. By completing this lesson, you now have a practical framework for designing reliable, transparent and maintainable AI systems. Whether you go on to build autonomous agent, task planners or tool enabler workflows, you are no longer just using AI, you are architecting it. And that matters. Continuous improvement, whether in AI operations or leadership, always begin with people who choose to learn. People like you. Let me share a little bit about myself. Who am I beyond this class. I'm Dimple Sangui, instructional designer, AI capability builder, corporate trainer, and a founder of AvisA Learning App. Over the years, I have worked with professionals and organization across industries, helping them build systems that are efficient, intelligent, and future ready. My goal has always been clear to make complex concept practical, to make powerful tools accessible, and to help individuals build capabilities that truly create impact. This class is just one part of that mission. But your journey does not end here, and I would love for us to stay connected. You can connect with me on Linden, where I regularly share insights on AI systems, continuous improvement, productivity, and leadership. You can also join my WhatsApp community for quick micro lessons and some practical tools and real world examples and instructor led programs, which are most of the time available for free. If you want structured deeper learning, explore the Aviza Learning app where you will find additional courses, templates, certification program, and guided learning challenges. Keep building, keep learning, keep experimenting, keep designing systems that think better and work smarter. Most importantly, keep investing in yourself. Thank you once again for being my students and being part of this learning journey. I will see you in the next class. Thank you.