Where to properly start with AI

A practical guide to get yourself quickly up to speed with generative AI and all that it can do.

Jul 08, 2025

Lately, I've been feeling like AI tech support for my colleagues, friends and family. The requests were relatively infrequent and pretty straightforward to start: Should I use ChatGPT or Claude? Where can I go to learn more about AI? What do I need to know about prompting? Most were quick and easy to solve with a few links or a short voice note.

Now the game has changed. I receive daily questions ranging from more complex topics such as opinions on AI concepts such as MCPs or Agents to how AI will impact knowledge work (that’s a podcast series, not a message). But I also get practical queries such as advice on using publicly available models for work or how to prompt for Deep Research. And I still get a lot of the basic 'how do I AI' type of chat coming my way.

Don't get me wrong, I LOVE a good AI chat. But, I am only one person who is trying to launch her own business and have some semblance of a life away from a laptop (spoiler: turns out this hard when you're a budding entrepreneur). So, as a time-poor individual who wants to support people get up to speed with AI, I have created a step-by-step guide to help.

Everyone should find something here, regardless of your level with AI. I have been living and breathing AI-related content for years now, and hope to pass along some of this wisdom to you. This guide points you to the resources to get started with the basics, and further your own learning and experience with AI.

Before we jump in though, waiting for formal training via your company is NO LONGER an option. You can't rely solely on 'trying Co-Pilot' now and again, or taking a single 'upskilling session' on AI to get you through. It's not enough. We no longer can expect someone to 'teach' us AI.

You are the only one that is responsible for your own learning and career development when it comes to AI. But hopefully, this guide will start you moving in the right direction.

How to use this guide

I really believe this moment is different. Much like the steam engine and electricity were the driving force behind the Industrial Revolution, AI will transform knowledge work in a way we have never experienced. But at a much faster pace, in ways we cannot yet predict.

Generative AI isn't programmed the way previous technology has been. This makes AI’s ever improving capabilities, and how it will change work, are hard to predict. That doesn't mean we can take a 'wait and see' approach. Rather, our understanding and use of AI should evolve and develop alongside the models.

We need to get comfortable operating in the ambiguity. If you are to thrive in the Age of AI, you must recognise this is an ongoing journey. There is no finish line. Just change.

The cycle of learning and applying Generative AI

Use this model to help you build your AI literacy and technical skills initially. Once the basics are sorted, you can also use this model to understand more advanced specific AI-concepts. Regardless of how you apply it, just know that your AI learning will follow all four stages and will continuously evolve as the frontier AI models get better every day.

Foundational knowledge

I argue everyone needs to understand the 'basics' when it comes to Generative AI. The challenge is that the 'basics' are not simple, and dedicated time is needed to get yourself up to speed, especially for us non-technology folk.

The good news is that the resources available today are plenty. I don't need to make any explainer content to help you with the foundations because it most certainly already exists.

Here are my recommendations to help you build foundational knowledge of Generative AI.

OpenAI Academy: Beginner 101 (videos are 5 – 20ish minutes)
Start here for short, beginner-friendly explainers that cover how AI models work, and ensure you have a grasp of basics concepts and practical examples of Generative AI use cases in specific contexts (i.e. education, non-profits).
AI Fluency: Foundations and Frameworks: Thinking about AI at work
(3 – 4 hours to complete the course)
This course provides an excellent framework that helps you collaborate with AI efficiently, effectively and safely through the 4-D’s: Delegation (thoughtfully deciding what work to do with AI versus yourself), Description (communicating clearly with AI systems), Discernment (evaluating AI outputs with a critical eye), and Diligence (ensuring you interact with AI responsibly).
Follow Andrej Karpathy and Nate B Jones: Two creators to help you sense-make AI (and there are many more)
These creators break down complex AI concepts into somewhat digestible explanations. Karpathy’s +3 hour Deep Dive into LLMs like ChatGPT provides technical depth for those ready to get into the details of these models, whilst Jones’ provides detailed overviews of latest developments in AI via his incredible Substack.
Deep dive into online courses: Google, Microsoft and Nvidia
These industry leaders offer (mostly) free, highly-specialised courses that combine theoretical knowledge with hands-on exercises, giving you both conceptual understanding and practical experience with their specific tools. Recommended if you want to go deep on the technical side of Generative AI.
Specialist, structured AI programmes.
Many big name universities are offering specialist, multi-week courses that give you an in-depth understanding of AI, often a hefty price tag. Some are targeted for general business or tailored for executives / leaders, or tailored to a particular field. Before you sign-up, how the course content will keep up with the ever-evolving changes in AI? Also, any programme should have a STRONG practical element where you're actually building something AI-related. Hard for us non-techies, but it does give you a sense of the advantages and limits of no-code tooling.

At a minimum and no matter what path you take, you should have a sufficient understanding of how these models work, key concepts like tokenisation, context windows, RAG, etc, what capabilities are at the 'frontier' (i.e. – Autonomous Agents, MCPs), and the ongoing potential issues (i.e. – hallucinations, memory, security, etc). This foundational knowledge will make all the difference for what you’ll get out of the next step: experimentation.

Experimentation

No other time has there been a technology where you must 'learn by doing'.

Everyone* should be paying for ‘Plus’ or ‘Pro’ access to a frontier model. The difference between 'free' and 'paid for' versions of these LLMs now is a chasm. Likewise, having an upgraded access to a paid-for version is an essential part of staying up to date as the models increase in capability very quickly.

Now, the question: which model should you subscribe to? ChatGPT, Gemini and Claude are all good options. The choice comes down to the capabilities you use most, as well as personal preference in terms of their interface and the outputs generated. If you are confused by that previous sentence, I would suggest you start with OpenAI's ChatGPT Plus.

For about $20-ish a month (whatever currency you're in), you get access to a range of models and functionality in which you need to experiment with. In the case of Chat, this looks like:

Learn the differences between models (i.e. – 4o for general knowledge vs o3 for complex reasoning vs. o1 mini high for data related problems). Each model is better at some things than others, and only by experimenting will you start to understand the difference.**
Try out different modalities. Chat can understand voice, pictures and video. It can read documents and transcribe notes. Chat can also produce these things based on what you ask it to do, like 'create a picture' representing X concept in the style of Y art (see below).
Experiment with functionality. Most frontier models have a 'deep research' function (aka – A research agent), where you can generate a comprehensive research report, complete with citations, from your prompt. Chat also has a bunch of different functions, like Tasks and GPTs, that are worth trying too.
Leverage integrations. You can link your Gmail, Google Docs, and a range of online tools such as Notion, GitHub, etc that will make working easier and allow you to generate from your own content and context.
Train a model for context. You can create a 'project' in Chat around a specific topic. You can provide 'context instructions' where you give specific guidance and understanding to help the model to generate more specific content. You can also upload a range of documents that too provide additional background information and help with contextual understanding.
Experiment as more is released. These models never stay still. AI companies are always releasing new features and capabilities, so it's not a one and done exercise. Keep trying new things and experimenting, every week.

You can use the same approach for other LLMs as well. It doesn't just have to be Chat. My personal go-to is Claude, but I subscribe to all 3 major frontier models, plus Perplexity. I also often play with DeepSeek and occasionally Llama to see how they are evolving and try out different functionality.

Generate an image for this Substack post in the style of Van Gogh.

Honing your AI skills

Through experimentation, you will start to develop skills that are specific to AI. How you apply these skills to your own work will be different and unique to the individual. But, we all need to be intentionally fostering these skills in our own work lives if we are going to use AI for true augmentation.

Here are some key AI skills you should be practicing and improving often if you are to make the most of AI:

Prompting. The art of communicating effectively with AI models to get the outputs you need. Good prompting involves providing clear context, specific instructions, and understanding how to iterate based desired output. AI generated output is only as good as the prompt provides.
Benchmarking. Testing different models against the same task to understand their relative strengths and weaknesses, either within the same platform or against each other. This helps you choose the right tool for each job and understand when one model might outperform another for your specific needs.
Validation. Critically reviewing AI outputs for accuracy, relevance, and potential hallucinations. This skill is crucial because whilst AI is powerful, it can confidently generate incorrect information, making diligent human oversight essential for ensuring the AI is truthful in its outputs.
Discernment and Judgement. Knowing when to use AI versus when human insight is irreplaceable. Right now, humans have better systems thinking, contextual understanding and are better at recognising the nuances AI might miss. This meta-skill helps you identify where AI adds genuine value versus where it falls short.
Intentional workflows. Designing your work processes to leverage AI where it makes sense whilst preserving the things that humans are uniquely good at, such as creativity. This means thoughtfully integrating AI into your existing practices as a place to start.
Preventing meta-cognitive laziness. Actively maintaining your own thinking and problem-solving abilities rather than defaulting to AI for everything. We need to be smart about when and where to use AI, and be mindful to ensure you don't lose the very skills that make you valuable.

Practice, regularly and often. Develop your own systems and processes to help you hone your AI skills in a structured way. Reflect on what you learn by doing. And keep going.

Continuous Learning

The AI landscape changes weekly, making continuous learning essential. It’s very easy to get overwhelmed by the volume of information out there, and the pace of change, so it’s essential to develop your own practices to keep up to date.

Here are some suggestions that will get you started:

Set aside dedicated time each week to explore new developments in AI in the news. Even better if you can use an AI to help you find and summarise information automatically.
Follow thought leaders who translate complex AI advances into practical insights, and participate in communities where practitioners share real-world applications. Ethan Mollick is a personal favourite.
Listen to podcasts during commutes and downtime. The Hard Fork Podcast in an excellent weekly that goes into the 2-3 biggest tech stories of the week.
Spend time offline. Time away from screens is so very important. It’s in the offline moments that we develop the uniquely human skills that will distinguish us from AI. Remember we have two distinct advantages over AI: we can connect to the physical world, and each other. So make time for both.

Make learning about AI a sustainable habit rather than a binge. When you integrate small experiments and micro-learning into your daily life is critical if you are to evolve and thrive with AI.