In the previous lesson, you saw the inherent "weirdness" of Generative AI. And in this lesson, we will see the most effective technique to combat it: Structured Prompting Frameworks. While this name sounds complicated, this lesson will completely demystify them.

Prompting Frameworks are all over the internet; Every thought leader, organization, and business school is creating a new one. OpenAI has recommended their best practices, and several other major tech companies. I have analyzed a dozen of these and I will share the simple meta-pattern behind all these frameworks.

--

OK, first what is a Prompting Framework? A Prompting Framework is a checklist, packaged as an acronym. You use it to start a new project with ChatGPT, or any other tool of your choice. For example, you can see 8 different popular frameworks on your screen now. I will demystify all these shortly.

Once you start your project in the correct direction, with the right context, then you can continue chatting with AI.

But if you start wrong, i.e. without a proper context, then there is a higher chance that hallucinations build up in your project.

If you use the right framework, then you get consistent, accurate, higher-quality responses. This is what structured prompting framework is all about.

--

There are 40+ prompting frameworks documented online. CARE, RISE, RTF, TAG, RACE, PACT, and dozens more like this. But here is the secret: they are all variations of the same four elements.

WHO: Who is doing the work? What role or expertise? This helps AI match to similar expert text in training data.

WHAT: What needs to be done? Clear goals help AI pattern better to completed tasks.

WHY: Why are we doing this? What is the goal or context? Context narrows pattern matching.

HOW: How should it be done? What format or steps? Explicit format gives clear structure to imitate.

Every framework is a remix of these four elements. Once you understand this pattern, you can use any framework or build your own. My recommendation is that you master these three Core Frameworks. They will cover ninety-five percent of your PM work.

Framework one: RTF. Role, Task, Format. This is the most popular format, and it will be your daily workhorse. It's simple, fast, versatile and proven. Role sets expertise context. Format gives structure to imitate. Use this when you need something fast and well-organized.

Framework two: CARE. Context, Action, Result, Example. Use this when you can reference to a successful EXAMPLE. CARE's superpower is the Example component. This is few-shot prompting technique built into a framework. This is exactly how consultants brief each other. You are showing AI concrete patterns to match. Research shows examples produce disproportionate quality improvements.

Framework three: RISE. Role, Input, Steps, Expectation. Use for multi-step processes. The Steps component creates chain-of-thought prompting. Each step generates more tokens, more pattern matching opportunities. RISE helps AI imitate problem-solving traces from it's training data.

--

Which framework when? Use this table to understand. If your task is quick and straightforward, use RTF, which should be your default. If you have an example of what good output looks like, use CARE. Leverage your example. If the task has multiple sequential steps, use RISE. Break it into a process. If there is a completely unique situation you are facing, then build your own framework. Simple as that!

--

As a bonus, here on the screen are 10 common project situations and the corresponding prompt framework that you can use to start your AI conversations. This is pretty simple to understand, please pause the video and read through if you want.

--

And now we come to the conclusion of this lesson. Here are the key takeaways:

1. A prompting framework is a checklist disguised as an acronym. It prevents you from forgetting critical prompt components.

2. All frameworks are variations of WHO-WHAT-WHY-HOW. Once you understand this meta-pattern, you can use any framework or you can build your own.

3. Master these 3 for immediate project work: RTF (daily default), CARE (when you have examples), RISE (multi-step processes).

4. Frameworks are training wheels like when you are learning to ride a bicycle. As you internalize the pattern, you will naturally include WHO-WHAT-WHY-HOW without needing the acronym.

5. Consistency beats perfection. Pick 2-3 frameworks you like and use them consistently rather than memorizing them.

"A picture speaks a thousand words". In this lesson, we will explore a wide range of Visual Generative AI tools available for your professional usage. All the tools that I showcase now, has free and paid access, so you should be immediately able to experiment everything that I show in this action-packed lesson.

On your screen now is my own project-management product called PlanShare. I will show how GenAI tools are used for different aspects, in 6 different cases. In the interest of saving time, I have created all of these, and you will find the exact prompts, attached with this lesson, which you can download and play with. Without further ado, here we go:

--

USE CASE 1: I want to create a Social Media Graphic for announcing a new feature to my existing users.

We will use Google's superb AI called Gemini. This is a full-powered competitor to ChatGPT.

Gemini includes "Nano Banana" which is one of the most hyped image creator.

I know it looks very simple, but this is a 5 minute iteration, and it's sufficient for a quick customer engagement, without having to expend time & money. Your customers also will be delighted when you take this simple extra step to show you care.

--

In the next USE CASE #2: I will show a Data Visualization for Customer Success Metrics, that will be used with my product team. I need to show a thirty percent reduction in delays, eighty-five percent satisfaction, ten thousand active registrations.

Notice that I have used an upcoming tool called Ideogram for this demo. You might have noticed ChatGPT is very prone to spelling mistakes in it's images, and that's the reason I have used this new tool, which is pretty good at rendering readable text and numbers. This is citical for data visualizations. So another tool in you can explore.

--

Moving on to USE CASE 3: I want to create a Blog Post Hero Image;

We return back to ChatGPT's DALL-E image component. Why? Because this is great at concept illustrations, abstract visuals, an artistic perspective. Here you see an abstract representation of AI and project timelines merging. Holographic Gantt charts. Once again, this is good enough for my blog.

--

For our next USE CASE #4: I want to create an Email Campaign Header Banner. And my weapon of choice is the crowd favorite: Canva.

Canva is now ready with their GenAI offerings, that you can use just like any other tool.

Please note: When you use Canva, it is also possible to directly edit into the design and make changes until you are happy.

And here is a special Tip: you can localize the language - see here I have the same message in Hindi! Actually, Google's Gemini is the best for localization, but other tools are also catching up, like Canva.

--

The next USE CASE #5, will be to build a Infographic for Competitor Comparison. Infographics are ppular because they convey a lot of information in a concise visual. I always do this with trusty old ChatGPT.

You can see the reason here: not only did ChatGPT generate an competitor comparison image, it also fetched real data from the respective websites and hilariously told me that I was not fully accurate. Use it as you will :-)

And here is pro Tip for you on the same topic: there is now more and more regulation for identification of AI generated visuals - images or animations. Gemini's Nano Banana is now loaded with SynthID watermarking, and C2PA content credentials so you can easily let people know image was generated with AI.

--

And now we come to final USE CASE #6: Here you can see I have generated a technically accurate flowchart for the schedule analyzer logic of PlanShare. I have used another GenAI tool called Mermaid.ai.

Such a flowchart would have taken me many hours to create manually. But now I can create and edit through text.

-----

I hope you enjoyed this lesson, and it should have gotten your creative juices flowing. Currently the real superpower of visual GenAI is MidJourney. This is incredibly powerful and can create professional advertising level images effortlessly. It is a premium/paid software that you can explore if your work requires this level of artistry.

All the use case prompts are attached with this lesson. You should take these prompts, improve upon them using the RISE prompt framework, add your project situation to it, create an image and post it in the Q&A section of this course. I will be delighted to see your creativity! Best of luck.

We will now conclude this lesson with some extra tips:

1. Nano Banana is my current favorite because it has access to Google's real-world images of people, places, and latest products. This is a game changer for educators, marketers, & product managers.

2. Many of these GenAI tools get the spellings in images wrong. You will have experienced this already. Again Nano Banana is the best tool with 12+ languages spell checked.

3. Except ChatGPT (which is dodgy), you can use all the other tools I showcased to get high resolution, production ready 4K images, with the aspect ratio of your choice.

In this lesson, we will discuss AI-Powered Research. We will use a very special tool for this lesson: Google's NotebookLM. There are some specific conditions which make NotebookLM as the perfect tool for research and analysis.

First, you want close to zero AI hallucination. Secondly, you want to prevent access to dubious internet content and focus only on your own premium content. And third you have your research input in a lot of different formats, like documents, images, videos, and sound files. This is a very common situation in our professional lives, and NotebookLM is the perfect tool for this. This can be life-changing.

--

OK, on your screen now is my own research workspace. While the NotebookLM interface might look more complicated that the other GenAI tools, it is actually quite intuitive once you get the hang of it.

<explain the screen>

1. On the topmost left of screen, you can the name of my notebook, you can edit it in-place. My notebook is titled "State of PM AI".

2. Observe that there are 3 main panels, the Sources panel, the chat panel, and the Studio panel.

3. The Sources panel is where you upload all your input files. In the free version, you are limited to 50 sources (or a total of 25 million words), which is quite substantial. Notice there is no file size limit, so you can load humongous data files too.

Notice that in my workspace, I have uploaded my personal research material. NotebookLM's policy states that this material and the generate outputs, will NEVER be used to reverse train the AI model. So, I would say this is relatively more secure than any of the other GenAI tools out there.

4. The central panel is where you chat and interact with the AI. This is where the magic happens, and it is just like the other tools we have seen so far, except that the research ONLY happens in the content that you have provided.

Notice that each and every finding has these numbers at the bottom. These are citations referring back to the original content.

5. The rightmost panel is for outputs of your research. This panel is rightly called the "Studio", because it can automatically create a variety of outputs - not just plain boring text! Audio, video, mindmap, reports etc. And many more coming in the future.

My absolute favorite is the "Audio overview", which creates multi-character podcast. Listen to this overview I created.

I can listen to these like my other podcasts. Infact, recently Spotify has used this feature recently to create their annual year wrapup, and made it available to millions of users worldwide;

--

Now here are some pro-tips, and pitfalls:

1. You can turn sources ON or OFF, as you wish, for individual queries, thereby controlling what will be included or not in the research answer. This is very helpful in my case, as I can iterate with multiple versions of my PMBOK guide.

2. Every time that you have a productive interaction with the AI, you should "Save to note", using this button. You will then be able to ADD this BACK to the sources, to continually keep building on your research material. This can be a critical feedback loop to your research. This is also a clever way to bypass the 50 sources limit, as you can combine results, and then remove the original sources.

3. you can customize the AI responses, by choosing the pre-defined styles, or you can custom build your own styles. I like to build my own Project/Product management vocabulary to my research notes.

4. You can share your notebook with anyone else of your choice like your team members. And they will be able to participate in your research.

5. Don't expect creativity in NotebookLM. This tool is for diligent critical thinking. You can take your research and then feed it into ChatGPT, or Gemini for creative outputs.

6. Be careful of what you add into your sources. If you give it garbage, you will be assured of garbage output.

--

And now we come to the conclusion of this lesson. NotebookLM is a game-changing tool for me personally, as an instructor and educator. The ability to grok huge amounts of information and convert it into reliable insights and wisdom is something I am finding irreplaceable.

In this incredible fast moving world that we find ourselves in, both personal and professional, our attention spans are getting limited and we have to sieve through enormous amounts content. NotebookLM brings back some semblance of sanity and confidence back to the output.

Speaking of outputs, no more spending 100s of frustrating hours creating slide decks and infographics for our teams. I hope you also enjoy your NotebookLM journey, and I will see you in the next lesson.

In this lesson, I will showcase a very powerful use case for GenAI: The ability to make sense of data.

In our everyday work, we are often bombarded with raw data.

And we have to make hard business decisions based on this raw data.

More often than not, the only way to work with this data, will require technical knowledge like database design, SQL, or some other form of writing scripts.

It can usually take you days together to make sense of even small & medium datasets. And moreover, you will be on your own, and you might be missing the hidden relationships in data.

--

On your screen now is a CHURN data for a telecom company. This is customer data, which includes people who are stopping our services. We have to quickly analyze who is leaving our services, how bad is the churn, and what we can do about it. As you can imagine, this kind of fast analysis can be mission critical for your business.

This is a pretty large dataset, with 7000+ rows and 23 columns. This exists as and anonymized comma-separated-text CSV file format. I will attach this dataset and the prompts used, with this lesson, so you can download and play with it if you want.

--

In the first step, I have directly uploaded the data CSV file onto ChatGPT. I will then ask it to make sure data is readable as a table. This is basic hygiene test.

ChatGPT also comes back with some of fields identified which can help us understand the churn data.

--

In the next prompt, I will ask for a fairly heavier analysis of the data. I want a 1-page executive summary of why some services are being canceled. I want to know some insights of what is happening in our spectrum of services.

Now I will open the insights summary found by ChatGPT.

Incredibly, the AI has returned back with some very interesting data - the top 3 churn drivers, with evidence to back up it's claims from the data.

There are also 3 recommended actions. Now the beauty of this is that LLMs have sufficient exposure to telecom business that these recommendations are not half bad! These can serve as the first stepping stone.

The real game-changer here is that all the time-consuming, error prone, technical stuff has been removed. The insights into the data is the icing on the cake. This can really catapult your data analysis tasks.

--

And now, I am going to drill down even deeper into the data. This is where the fun really starts. I will ask AI to compare the highest and lowest churn segments.

It has returned with a bunch of analysis; But here it gives me the real business insight for highest and lowest churn.

--

In the final step, I want chatgpt to convert all this analysis into actionable insights. What are 3 actions per segment that we can consider? what can be the KPIs for each action? and how can all this be tracked weekly?

So the real lesson here is not “AI can make charts.”

The real lesson is: AI can help you move from raw data → clear patterns → decision-ready actions, without spending days together in spreadsheets and databases.

And as a professional, that is what you need: fast clarity, fast decisions.

--

I have spent days and months analyzing different datasets as a project and product manager, using SQL server, Excel, Power BI etc, and it really blows my mind that we can query data in a natural language. And even more so that the AI is able to go further than what we even ask, and gather insights from under the rock, so to speak.

So, in conclusion, I urge you to replicate this hands-on lesson with the attached material. And also, try out these capabilities in your worklife. But be careful and follow your company data policies. See you in the next lesson.

<intro>

This is the right time to talk about some very interesting, and pertinent topics: copyright, ethical considerations, and regulations. Especially for those of you building new AI products and services.

<insight:1 Bias is default>

Unless you have been living under a rock, you know the Internet is full of bias.

This is because we humans are full of bias.

What happens when you train AI from literally everything on the internet?

Naturally, you can expect some bias will be inherent in GenAI - in text or images or video.

You have to review the AI outputs with a critical eye.

BTW, what exactly is a bias? The dictionary says: Bias is a partiality that prevents objective consideration of an issue or situation. It is an unfair influence being exerted.

Now I will share a real world example of how bias can exist in AI: In 2023, a tutoring company's recruiting software automatically rejected older job applicants without human review, violating age discrimination law. The company settled with a fine in the first-ever AI hiring discrimination case, paying damages and implementing new anti-discrimination policies.

Since then, there have been several complaints that have come up.

<insight:2 Copyright>

New York Times has sued Microsoft / OpenAI for major copyright infringement lawsuit over AI training.

Some AI companies have scraped the entire internet, and trained on it without permission. Publishers and artists are suing. The defense: We're learning patterns, not republishing.

The Courts haven't decided yet. Different countries, different rules. UK and Japan allow it. EU requires transparency. US is figuring it out through litigation. In India the legal machinery has started moving.

The bottom line for now is: if you are working with copyrighted material, ensure that your generated outputs do not trickle back into the AI's training data. You will be held responsible for this.

<insight:3 Regulation>

Regulation is fragmenting.

The EU has the strictest rules. AI Act, will be in full enforcement from August 2026. High-risk AI will be in hiring, fiscal lending; healthcare requires human oversight, bias testing, and explainability. There can be tangible penalties if you are on the wrong side of regulations.

The US is litigation-driven, with no federal laws yet. Courts are setting precedent through lawsuits.

In India, NASSCOM guidelines drive Industry self-governance.

This matters if you build global products, you should then design for the strictest jurisdiction. That's usually the EU.

This is real. Companies are restructuring how they build AI because thirty-five million euro fines could end the business.

<best practices>

If you are building AI products, or shipping out generated content, you should have asked some questions:

- Where did the training data come from? Are there any Copyright risk, bias risk, or privacy risks.

- Could this create biased outcomes? Have you tested on different demographics?

- Can you explain how your product makes decisions? High-stakes decisions can not have black box design.

- Is there human oversight in your AI processes? Can someone review and override?

- Do users know they're interacting with AI?

The reality is that some bias is inherent in GenAI.

--

And now we come to the conclusion of this lesson.

Just like any other engineering project in the world, there are trade-offs when you work with AI.

The problem really is that this is all so new and moving at a break-neck speed.

Some of the tools that we use today, might vanish in a bubble tomorrow.

But the great AI shift has already happened, and there is no going back.

There are no perfect answers, but there can be conscious trade-offs.

And that is what we should work towards.

I will now end this lesson and we will return to our skill building in the next one! See you there.

In this lesson, I will introduce the exciting future of desktop computing: Claude Cowork.

This is a desktop application, that can connect to your desktop, access your file system, connect to any of your business applications, and you can delegate work to it. This application brings you the power of programming, even if you are a non-programmer.

--

A couple of caveats before I go further into this lesson: This app is available for paid subscribers, I have the Pro plan, which costs approx $17 USD/month. Second caveat is that this is absolutely fresh off the oven. It was released a couple of weeks back, and it has completely blown me away. You can build a 20-member agentic team with this tool if you know what you are doing. I am still new to this tool but I will show you a few things that will reveal what this tool is capable of.

--

Let's take a moment to understand the UI.

Notice there are 3 tabs at the very top, and they bring together Chat (which you will already be familiar), then Cowork (which is what this lesson is about), and also the notorius Claude Code - where you can program code.

You can work seamlessly between all three offerings. How cool is that?

--

Now let's focus on Cowork.

Claude Cowork is basically an AI agent that runs on your desktop and executes multi-step tasks autonomously.

If you are of a certain age, like me, you will remember the Agents from the movie Matrix.

You can say this is the first version of those agents :-)

Unlike Chat (where you ask questions and get answers), Cowork does actual work, organizing your files, creating spreadsheets, generating reports from your data, while you can step away from your computer and go to the gym.

You can synthesize data from multiple sources, you can format deliverables from raw data and so on. In my early programming career, I have written dozens of programs that would do things like this.

--

OK, before I show the REAL game-changer, let me mention the impact Claude Cowork has done to the market. After Claude Cowork was released, the major Enterprise Software company stocks dropped more than $200 billion These were the companies whose core functionality overlapped with what Claude Cowork's desktop AI can automate. And Microsoft has swiftly responded with their own Copilot Cowork.

OK, what exactly is causing this disruption?

Notice here at the bottom: "Customize with plugins". The magic happens when you enhance your app with plugins.

Cowork ships with a few dozen plugins, right out of the box. But your company can build it's own plugins for specific roles.

The plugin that most companies can use immediately, straight away with least friction is the Customer Support function. This can be a miracle for small business owners like me. And at the same time, it can save millions for large organizations also. This plugin can triage tickets, draft responses, escalate issues, provide it can provide voice support (which is slowly getting indistinguishable from human voice).

The concept of a plugin is very simple: It is a predefined set of (Prompts + Context + Configurations + System Instructions) all rolled into a package. But when taken together, they can be incredible.

You can build agents for ANY business function: HR, Legal, Customer Support, Engineering, Finance and so on. For this to be actually functional, you will have to take the base template and customize it to your business requirements.

And THIS is what the future will look like, whether we like it or not.

Allow me a couple of minutes to ponder on this tool: How best can we adopt these new tools & new possibilities? How can we improve our services, our offerings to the world, how best can we improve our skills? That will be the key question, both from a personal and organizational perspective.

I have not built any demo for this lesson, because this needs a new course by itself. But, I wanted to show you what is coming up ahead in the very short future.

--

And with that I will conclude this lesson.

Claude Cowork is mindblowing because you can build agent-grade automation, and parallel workflows - even as a non-programmer. You can possibly build a 20-member agentic team working for you, non-stop 24/7.

On the flip-side, this is still bleeding edge technology put in your hands. There are fast evolving security concerns, Trust issues, and I am sure a bunch of reality checks will come ahead.

But, this is the future of work, without any hype. Tech companies like Spotify have already adopted Cowork in areas like code migration. I will build more detailed tutorials for this in the future, but for now, I will see you in the next lesson!