FIRST OF ALL – INTRODUCTION

For a technology widely acclaimed as an assistant or the perfect productivity hack, the use of AI can sometimes feel very strenuous and ironically inefficient.

“Garbage In, Garbage Out,” they say. So, you learn as much prompt engineering as possible. When the time comes to use the learned principles, you try your best. You spend minutes or even hours typing out long, detailed instructions like, “Act as a Gen-Z Nigerian social media manager…” or “Please summarize this report in bullet points, keeping the tone professional and accessible to an undergraduate.”

You hit “Send” and your AI gives you responses that make you hiss and ask: “Wetin be this?” It got something wrong, or missed an important detail, or you omitted something crucial in the prompt. Or in the worst scenario possible, it hallucinates. So you wrestle with it in frustration. You reiterate, craft your prompt again, or on rare occasions, settle for the flawed output.

Three days later, you have to repeat the process, so you do it again. That’s the digital equivalent of digging up a well and filling it back up. E no too make sense!

What if I told you that there is a gem tucked away inside Google Gemini that lets you hire a specialist—a dedicated expert trained by you—to handle all that repetitive work, permanently, and still deliver maximum quality?

This way, you don’t have to remember any perfect 10-line prompt. You simply have to give your AI a permanent yet editable identity. You create the prompt ONCE, and you’re good to go forever. I’m talking about Gems.

What Are Gems?

In simple terms, a Gem is a customized version of the Gemini model that remembers a specific, detailed set of instructions forever.

An analogy that best differentiates standard AI prompting from using a Gem is different employment styles. Think of standard AI prompting like hiring a freelancer for a single gig. You explain the project, they deliver, and the next week, you have to explain the project again to someone new. Double wahala!

A Gem, on the other hand, is like hiring a full-time, salaried employee with a crystal-clear job description. You define their role—their system prompt—once, and they keep that role as their foundational identity for every single task you give them.

This initial system prompt sets the rules:

1. The Persona (Who It Is): “You are a witty, authoritative, and slightly sarcastic Nigerian food critic based in Lagos…”

2. The Task/Goal (What It Should Do): “Your primary task is to take a set of raw, unpolished tasting notes (provided by the user) and transform them into a compelling, polished, and shareable restaurant review.”

3. The Format (How to Deliver): “Always deliver the output in two distinct sections…”

4. The Limitations (What It Should Not Do): An example is: “DO NOT use generic Western-centric food terms. Instead, use evocative local expressions…”

5. Its Knowledge Base: “Reference the attached Brand Style Guide PDF for all tone and color palette decisions.” This is important, but optional. For example, if you’re a content writer, you can attach your previous writing projects so that your Gem can understand your writing style better or gain more context.

Creating a Gem like this essentially sets up a persistent, long-context-memory AI agent. It saves your time, reduces errors, and dramatically improves output consistency.

KINI ANFAANI? WHAT DOES IT DO?

The application of Gems for AI users choke! It can do so many things, especially repetitive and time-consuming tasks like:

1. Content Creation: Gems guarantee brand consistency and format adherence across all channels. They permanently embed your specific voice, tone, and required formatting (e.g., use of local slang, hashtag rules, image description style) into every draft.

2. Business Tasks: For businesses, Gems can help you enforce structural routine. You define the exact framework (e.g., PESTLE, SWOT, monthly reporting template), and the Gem ensures all reports, summaries, and meeting minutes strictly follow that template every time.

3. Personal and Study: Gems can serve as a specialized tutor or planner. That thing wey you no understand because your lecturer dey use big big English? Gems can help you break it down welewele, based on the same analogies and learning steps you defined.

4. Technical Tasks: As a developer, you can use Gems to screen your codes and solve technical issues. It can eliminate variability in technical output by permanently remembering style guides, documentation requirements, and preferred coding languages.

Ati bee bee lo! (And so on, and so forth!)

How It Works (+ Free Example)

A Gem is the perfect example of Small body, big engine. It has a radically simple workflow and a surprisingly massive impact.

First, you have a choice: you can start by picking a ready-to-use expert from Google’s gallery of Premade Gems (like a Career Guide or a Trip Planner) and customize it, or you can build a Gem from scratch.

If you choose to build your own, the process is straightforward:

1. You Define the Prompt: You go into the Gems builder and write the comprehensive system prompt detailing the agent’s identity, rules, tone, and formatting.

2. The Gem is Born: You save this persona and give it a name. We’ll name this one “Agba Food Critique (Senior Food Critique).”

   

   Oluwakorede X Google Gems (Source: Screenshot by Oluwakorede)

3. Repetition is Eliminated: From that moment on, whenever you open this “Agba Food Critique,” you don’t need to re-type the role. You simply paste your data, or tell it your query, like “This is this week’s food review.”

4. Long-Term Memory: Because the system prompt is always active, the Gem retains context and behavior across weeks, eliminating the frustrating need to remind the AI who it is and how it should speak. It’s the difference between coaching a beginner and delegating to a pro.

EXAMPLE: AGBA FOOD CRITIQUE

QUERY

OUTCOME

WHO YOU EPP? WHO’S IT FOR?

Gems are designed for anyone who performs repetitive, defined tasks.

• The Busy Professional: You can create a “Weekly Summary Gem” to instantly structure your meeting notes into a cohesive summary for your team every Monday morning.

• The Marketing Manager: Establish a “SEO Blog Writer Gem” to ensure all blog drafts consistently meet specific keyword density, formatting, and word count requirements.

• The Content Creator : Establish a “Brand Voice Adaptor Gem” to ensure all outputs adhere to the same brand standards and key messaging.

• The Student : Build a “Concept Simplifier Gem” that automatically processes dense reading materials or lecture transcripts and formats the output into simplified, bulleted notes or flashcards, ready for quick revision. (Note: As a student, you can maximize this tool even more with the enablement of your free Google AI Pro subscription)

• The Consultant: Design a “Client Proposal Drafter Gem” that embeds their unique value proposition and brand voice into every new proposal, regardless of the prompt.

PROs and CONs

Why You Should Consider Using Gems

1. Persistent Agent: Gems possess long-term memory for specific tasks, unlike standard chats.

2. Reduces Repetition: They eliminate the need to copy/paste or craft complex instructions daily.

3. Very Large Context: Compared to other popular models, Gems have very large context windows. This helps them to handle extensive data for specialized workflows (e.g., summarizing massive documents).

4. Specialized Workflows: You can have different Gems doing different things effectively for you. This is ideal for creating high-fidelity, highly consistent output for niche needs, as opposed to using one model for different queries in the same chat!

1. Easy Onboarding: The simple interface makes it quick for beginners to create their first specialized AI.

2. Knowledge Uploads: Can reference uploaded files (e.g., style guides, past work, documents) for hyper-specific context.

3. Sharable Agents: Gems can be shared with colleagues, friends, or teams, massively accelerating team efficiency and standardization. You can share a standard Gem with your team, granting either Viewer access (to use only) or Editor access (to refine the Gem’s instructions) for uniform, high-quality results.

WAIT, Just Before You Jump In… (Shine Ya Eyes)

1. Requires Clear Design: Just like every tool, it is only as good as the hand that wields it. The quality of the Gem depends entirely on the clarity and detail of your initial system-prompt instructions. If you get it wrong, it will too.

2. Platform Dependency: Currently, the full feature set is integrated deeply within the Gemini ecosystem, meaning you gats commit to the platform. Wahala for who no like Gemini before!

3. Lesser-Known Feature: Since it’s newer, the community templates and shared best practices are not as widespread as general prompting advice.

4. Instruction Sensitivity: If the initial instructions are contradictory or ambiguous, the output accuracy will suffer; garbage in, garbage out.

KOKO OF THE MATTER (Bottom Line)

Artificial Intelligence was invented to enhance human efficiency, but it isn’t efficiency if we spend hours battling with the technology. So, “I sabi use AI, I sabi use AI” no be by mouth! Is it really making you efficient?

With Gems (when used well), you don’t have to ask yourself this question. It mirrors everything a typical African AI user looks for: Resourcefulness, efficiency, and building something that lasts. Why waste time rebuilding the instruction set for your AI every single day when you can build a system that runs itself?

I challenge you this week: Don’t just prompt Gemini — or any other AI. Spend one hour crafting a single, perfect Gem based on the most annoying, repetitive task in your professional or creative life.

You’ll realize that the true power of AI is not in the conversation you have today, but in the systems you build to run tomorrow.

FIRST OF ALL – INTRODUCTION

ElevenLabs again? Yes o, e choke!

Last week, we reviewed ElevenLabs, focusing on its exceptional text-to-speech quality and voice cloning features. But that’s not all there is to it. The company is now pioneering a capability that moves beyond generating content to automating complex action: the Agentic Workflow.

This feature represents a breakthrough for African professionals, enabling them to transform their high-quality voices into intelligent, autonomous, and multilingual employees. These are not chatbots that just read a script; these are AI Agents that can think, plan, and execute real-world tasks.

What is the Agentic Workflow?

The Agentic Workflow is an advanced automation system that allows an AI Agent to function like a proactive, decision-making employee powered by realistic voice technology.

It moves beyond simple Q&A by implementing a dynamic, goal-oriented process, making the agent not just a responder, but a doer.

KINI ANFAANI? WHAT DOES IT DO?

1. Autonomous Action Orchestration (The “Agentic” Core): You know that thing that all of us like to put on our resume: “Ability to work without supervision”? That is exactly one of the things that ElevenLabs brings to the table. It can break down a complex, multi-step goal (like “process a refund and file a ticket“) into a sequence of smaller steps, execute them without human instruction, and coordinate the flow of the entire conversation.

   This capability separates the Agent from a standard chatbot. It enables African businesses to automate entire multi-stage operational processes—not just the initial step—reducing human handoffs and ensuring that complex queries are resolved from start to finish.

2. Real-Time Tool Calling and Integration: This is the intrinsic function that allows the ElevenLabs Agent to integrate and interact with external enterprise systems (databases, CRMs, APIs, payment gateways) in real-time during a conversation.

   This is crucial for value creation on our continent. By granting the agent access to tools, it moves from being an information source to a service executor, capable of checking a mobile money balance, updating an account address, or flagging a fraudulent transaction as a customer speaks (functions vary based on industry).

3. High-Fidelity, Low-Latency Voice Synthesis (TTS): This refers to the platform’s core Text-to-Speech technology, specifically optimised for real-time, bidirectional voice conversation. It uses low-latency models to ensure the conversation feels natural, with a human-like pace and turn-taking.

   This ensures that service is not only fast but also relatable. It allows the agent to maintain high emotional fidelity and brand consistency, crucial for building customer trust over the phone, especially when the service is delivered in local languages or cloned voices.

4. Multilingual and Accent Adaptation: The model can automatically detect, understand, and fluently speak in multiple languages and regional accents within a single conversation (if you activate it), often using the superior multilingual models. This is a democratising force for African services. A single deployed agent can serve customers speaking English, French, Swahili, or Igbo, breaking down language barriers and allowing for continental-scale service deployment without language-specific agents.

5. Multi-Agent Workflow Orchestration: The model can also route a conversation seamlessly between multiple specialised AI agents (Subagents) or to a human. For instance, passing a query from a “Front Desk Assistant Agent” to a “Back Door Specialist Agent” or a human supervisor.

   This prevents the main agent from becoming overwhelmed or insecure. It allows you to scope context and tool access—a Front Desk Assistant Agent might only have access to a FAQ, while a Back Door Specialist Agent gets access to the payment API, dramatically improving security and accuracy.

Ready to Build Your Agent?

Now that you understand the power of autonomous action, the next step is deployment. Below is a tutorial that can guide you in setting up your very first ElevenLabs agent, be it for fun or for professional purposes. When you are ready, use our affiliate link to sign up and get a 50% discount on your subscription plan!

WHO THIS THING HELP? WHO’S IT FOR?

This technology moves beyond the scope of a single department. It is relevant to anyone needing to process information, make decisions, and take action autonomously:

• Businesses (Startups to Enterprise): Banks, Telecommunications, Service providers, E-commerce, and Airlines can use it to automate customer support, lowering the cost per resolution.

• Non-Profits (NGOs and Foundations): The ElevenLabs Agent can be used by this demographic for beneficiary vetting, donation management, and running large-scale multilingual public information campaigns.

• Government and Public Sector: The government can utilize the model for efficient citizen service, emergency mitigation, utility bill management, and automated public works ticket logging.

• Developers and AI Builders: Developers can use the robust API and visual workflow editor to build custom, high-stakes conversational applications, often referred to as “voice-first applications.”

See Agentic AI in Action

You’ve seen the potential. Since our team is also leveraging this breakthrough technology, you can see how we’ve implemented these powerful agentic workflows for real-world clients. Check out our custom Agentic Workflow solutions on our website!

Tip: You will find it at the bottom right corner of your screen.

PROs and CONs

Why You Should Consider ElevenLabs Agentic Workflow

1. Unmatched Efficiency: Handles high volumes of complex, multi-step tasks $24/7$, drastically reducing wait times and operational costs.

2. Scalable Empathy: Uses hyper-realistic, often cloned or branded voices that feel human, preserving brand trust while scaling capacity.

3. Multilingual Power: Agents can be instantly configured to support major languages and local dialects, ensuring service delivery is inclusive and culturally relevant.

4. Goal-Oriented Action: Moves past simple FAQs; the AI is programmed to achieve an objective (e.g., “Reset the password and send confirmation”), not just provide information.

WAIT, Just Before You Jump In… (Shine Ya Eyes)

1. Connectivity Requirements: The real-time nature demands a stable, low-latency internet connection, which can be challenging in certain rural or high-traffic areas.

2. Pricing Complexity: The credit-based system, charged by the minute of conversation, requires careful monitoring, as high volumes can lead to unpredictable overage costs.

3. Integration Reliance: The true “agentic” power only works if it can connect to your internal systems via APIs. Companies with legacy, closed systems will struggle with implementation.

4. Ethical and Trust Issues: It requires robust guardrails and explicit human handoff protocols to prevent “runaway” AI behaviour and ensure customer data security in high-stakes transactions.

Koko of The Matter (Bottom Line)

Agentic Workflows are not just about business efficiency; they are a development accelerator. By automating administrative, educational, and informational barriers in a human voice, ElevenLabs enables institutions to scale empathy, access, and governance in ways previously impossible. The true power lies in deploying these agents to serve the 24/7 needs of the common citizen, irrespective of language or location. A foundational tool for building the next generation of voice-first services in Africa.

Every now and then, a tool comes along that doesn’t just feel new, but feels right. I’ve spent years watching technology try to keep up with the beautifully diverse ways we learn. This week, I had a real ‘full circle’ moment while giving another test-drive on Google’s NotebookLM, and I knew I had to share why. Here’s my take on the AI tool that finally feels like it gets us.

As an educator, we talk about VARK learning preferences—visual, audio, reading/writing, and kinaesthetic—and for years, we’ve had to patch together different teaching methods and tools to enable learning for everyone. But with Google’s NotebookLM, I believe we’re seeing a tool that has finally come full circle, elegantly catering to all these learning modes in one powerful package.

So, kini big deal with another AI tool? The magic of NotebookLM is that it’s your own personal research assistant, grounded in your sources. It’s built on a concept called Retrieval-Augmented Generation (RAG), which is a fancy way of saying the AI answers your questions based on the specific documents or sources you give it, not the entire internet. This prevents it from sòsọ kú sò (hallucinating) and keeps the knowledge relevant to you. I call it a “public RAG”—a tool that gives everyone the power to build their own specialized AI, whether you’re a student in Lagos, a professional in Nairobi, or a researcher anywhere in between.

The real genius lies in the Studio, which is where the tool truly adapts to how you learn best. Let’s go with the whole learning preference thing – hopefuly, you get why this touches everyone.

• For the Visual Learner: The recently launched Video Overview, think of it as an AI-generated PowerPoint presentation, complete with an audio narrative, sometimes like a podcast style, other times its just a single speaker. From my own tests, I’ve seen a consistent structure: it creates a cover slide, then moves through different layouts—some with charts like bar graphs, others with tables and It can even generate timelines.

  This feature seems to be an evolution of concepts from Google Vids, another AI tool for creating presentations, showing how Google is building a connected ecosystem of products. What truly puts the learner in the driver’s seat is the ability to customize it. You can prompt the AI to focus the entire video on a specific topic or theme from your sources, which is incredibly powerful for narrowing down a large amount of information.

• For the Reading & Writing Learner: This is the foundation of NotebookLM. The chat interface allows for a classic Q&A, while the ability to add your own Notes lets you type out your thoughts and insights, adding them right back into the AI’s context. And of course, you can always upload notes you’ve handwritten and digitised.

  

   

  NotebookLM Add Note

  But it goes further with the Reports feature. Here, the AI can automatically generate study guides, briefing documents, FAQs, and even project timelines from your sources. This is perfect for anyone who needs structured, written material to study from.

  

   

  NotebookLM Reports

• For the Auditory Learner: Forget just reading text. NotebookLM creates an Audio Overview, turning your complex research papers, articles, or textbook chapters into a dynamic, podcast-style conversation between two AI voices. You can even prompt it to focus the discussion on specific themes from your sources. It’s like having personal tutors breaking down ideas for you on your way to Badagry. The best part? The interactive feature allows you to jump into the conversation, ask questions, and get real-time answers, as if you were a guest on the podcast. My guess is that this is likely powered by Google’s efficient Flash models, which are designed for rapid responses. It’s a brilliant application of a true multimodal system—one trained from the ground up to understand text, audio, and images seamlessly.

  

   

  Audio Overview

• For the Kinaesthetic Learner (The “Doer”): That interactive Mind Map is perfect for learners who need to engage and tinker. You can click, expand, and dive into different concepts, actively navigating your knowledge instead of passively consuming it. The entire platform encourages this hands-on exploration. Them finish work here! You can have the Audio Overview playing in the background while you navigate the mind map. This creates a powerful, multi-sensory learning experience where you can hear the concepts explained conversationally while you visually and physically trace the connections between them. It’s a dynamic way to see the big picture and dive into the details simultaneously.

  

   

  NotebookLM Mind map with Audio Overview

Collaboration and a Connected Ecosystem

Beyond individual learning, NotebookLM is built for teamwork. You can easily share your entire notebook with a study group or colleagues. You have two options: share the full notebook, giving collaborators access to all your sources and the chat history, or share a “chat-only” version, allowing them to ask questions based on your sources without seeing the documents themselves. This makes it an incredibly flexible tool for group projects and shared research.

As you can see, it touches on every possible way we consume and engage with knowledge, putting the learner firmly in the driver’s seat. The collaboration feature—allowing you to share your notebook with a study group—is just the icing on the cake.

Google already has a product called Illuminate that lets users choose from different host voices for audio summaries. Bringing that level of choice to NotebookLM—or even allowing us to clone our own voices—would be a massive step toward accessibility for millions of learners.

Rooms for Improvement

However, the tool isn’t perfect. I’ve noticed the audio in video overviews can sometimes have sharp, abrupt cuts – honestly, not a big deal for now.

But my biggest thought for improvement comes from an African perspective. For this tool to be truly transformative across the continent, it needs localization to address a major barrier to adoption: accents! (maybe you get it, but not all Africans watch American films). The default American accents in the audio and video overviews can be difficult for many learners to connect with. My key recommendations are to provide users with the ability to select from a library of pre-built local accents (like Nigerian, Kenyan, or Ghanaian) for the narrations.

Google’s other products, like Illuminate, already offer a choice of host voices, proving this is possible (although no African voices). An even more powerful step would be to allow users to clone their own voices, creating a truly personalized and accessible learning experience. This would be a massive step toward making the tool feel natural and relatable for millions of learners across Africa.

But for power users like me or ITKs there might be a problem of source limits (50 for free, 300 for Pro). Although it makes sense to maintain accuracy becasue of context window issues (SKIP THIS IF YOU SABI CONTEXT WINDOW IN AI — it’s a necessary technical safeguard. This is because of what’s known as the AI’s “context window”—think of it as the model’s short-term memory. There’s a finite amount of information it can “look at” all at once to answer a question accurately. The source limit ensures that all your documents can fit within this window, preventing the AI from “forgetting” information and making errors. While Google’s models boast massive context windows (up to 10 million tokens), this cap is a deliberate choice to guarantee the tool remains reliable and doesn’t hallucinate—which is the whole point of a RAG-based system.), serious researchers might hit that ceiling quickly.

KINI BIG DEAL (Why Does this matter)

For the first time, a single AI tool has finally come “full circle” by successfully catering to all major learning styles—audio, visual, reading/writing, and kinesthetic—in one integrated package – ALL FEATURES FOR FREE

• It’s Your Personal, Grounded AI: Unlike general chatbots that can make things up, NotebookLM is grounded only in the sources you provide. This makes it a reliable and accurate research assistant that gives you personalized answers without hallucinating.

• It’s a True Pedagogical Partner: It doesn’t just give you information; it transforms that information into the format that works best for you. It can turn the same set of documents into a podcast (Audio Overview), a presentation (Video Overview), an interactive diagram (Mind Map), or a written report. It respects and adapts to the individual learner, which is a massive step forward for personalized education.

What a time to be a learner. Ultimately, the best tool is the one that adapts to you, not the other way around. NotebookLM is a powerful step in that direction. It’s a reminder that the future of learning isn’t just about smarter tech, but about technology that allows you to still be you!

The only question left is: watin you won sabi? Start here: Google’s NotebookLM

I’d love to hear your thoughts. Have you tried NotebookLM? What has your experience been? Hit reply and let me know.

Until next time, stay curious.

For a while now, educators and parents have been asking the same question: with an AI that can answer any homework in seconds, how do we ensure our children are actually learning? OpenAI, the company that put this powerful technology (ChatGPT) in everyone’s hands, has just introduced a possible answer.

Let’s be honest—Bingo is long out of the cage! We know that simply dumping an assignment prompt into a chatbot bypasses real learning; a recent MIT study even proved that true mastery requires active engagement. This is why OpenAI’s new Study Mode is such a big deal.

So, how does it work? Instead of just handing you the answer, Study Mode acts as a personal tutor. Based on their announcement, here’s how it works:

• It Asks Questions Back: It uses interactive, Socratic-style prompts to guide you, helping you think critically and arrive at the answer on your own.

• It Breaks It Down: It presents information in smaller, scaffolded sections, making complex topics easier to digest and showing you how ideas connect.

• It Quizzes You: It includes built-in knowledge checks and personalised feedback to help you track your progress and make sure the information gums to your brain! (hopefully).

• It Adapts to You: The mode can adjust to your skill level, providing the right amount of support to keep you challenged but not overwhelmed.

  Learn More about Study mode here

And it’s interesting to see that OpenAI isn’t alone in this. This feature is giving the same vibe as Google’s NotebookLM, another tool built with a similar spirit. While they work differently—NotebookLM acts as a personalised expert on the documents you upload (like lecture notes or articles), while Study Mode tutors you on general topics—the core philosophy is the same. This shows a bigger, more encouraging trend: the major AI labs are finally building tools for genuine comprehension.

What does this mean for a teacher in Abuja or a parent in Accra? It means we finally have a version of this tool we can champion. While it won’t magically stop students from looking for shortcuts, it’s a much-needed shift toward responsible, pedagogical design.

And this isn’t just for students in school uniforms. Think about the entrepreneur in Lagos trying to master digital marketing, the Aba man looking to understand one engine he just imported, or even the seasoned professional wanting to get up to speed on data analysis. As they say, we are all students at every stage of life. Study Mode offers a powerful way to break down complex professional skills into manageable steps, helping you learn at your own pace without the pressure of a formal classroom. It’s a tool for anyone with the curiosity to learn something new.

Kini Big Deal?

Looking ahead, I can even see a future where ChatGPT (or any of these Chatbots)—perhaps with the rumoured GPT-5—intuitively activates Study Mode when it detects a student’s query. Imagine an AI that creates “intentional friction” to guide users away from ojoro (cheating) and toward genuine understanding.

This move signals that OpenAI is listening not just to businesses and creatives, but also to the needs of students. For us—educators, parents, and even professionals looking to master new skills—the call to action is clear: don’t sleep on this.

We need to get our hands on Study Mode and explore its potential. The more familiar we become, the better equipped we will be to guide the young people in our communities. Oya, click here to try it out now now

Looking forward to you sharing how you are using Study Mode. Share this with your guys.