The One With The AI Skills
You have to learn about AI. This blog is a must-read. It might be a bit technical or confusing, but the world is changing, and this is the new norm.
Hi guys, welcome back to yet another episode of Random Thoughts with Takura. There has been so much noise about AI that for a moment, I felt like this wave was skipping me. New tools every hour. Debates on every social platform. Fancy words all over: AI , prediction , tokens, transformers, models etc. But I just wanted something simple and useful. So I spent last week drafting this blog to give you what you need to feel current without feeling lost. DESPITE POPULAR BELIEF THERE ARE ONLY THREE AI SKILLS YOU NEED TO LEARN AND I WILL DIVE INTO THEM IN THIS BLOG. I will keep it clean. I will show you the tools I use every day. I will give you the principle that ties it all together so you can start using AI in your life.
The first skill is vibe coding. Vibe coding is how I use AI to build websites and apps that actually work and actually ship. I do not make teams anymore or try to write the code myself. I start with a simple plan and let the tools help me move. My stack is honestly light. Firstly, I refine my idea in ChatGPT until it sounds better, and I am definitely impressed by what I am saying. I then open Lovable and let it build the first working version of that idea. Lovable gives me a fast draft that I can click and feel. Its very simple to use, I just give it a prompt and it builds the website. I move that draft into Cursor and refine the app until it is clean and very easy to use. Cursor lets me type whatever I want in English, and then it converts it into code. Supabase takes care of my database so I can store users and lists without stress. I get a domain from Namecheap - so that's something like thecubicleafrca.com. And finally, I host it for free on Netifly. This whole process I just described, as simple as it is , that is how The Cubicle moved from idea to a real thing people use. That is how HouseHand got its first version in front of real people. It is not magic. It is vibecoding.
Ask. Draft. Refine. Connect. Deploy. Share. Fix. Repeat.
Trueman launched Finora with this exact script. You can sign up for his app right now while it's free - cause he might start charging soon.. He did not wait for a team or a developer. He picked the tools above, saw that he needed a solution to tracking his finances, and just like that finora.cash was born. Yvone built her organisation's website in the same way. She wrote a clear ask into Lovable, clicked through the first version, improved it, and improved it, now anyone who searches the share organization can find what she is doing on theshareorganization.org. She published and learned what to fix by watching people use it. That is the flow I believe in. You do not need a developer to start. You need a decision and a day on the calendar. AI makes the first draft fast. Your taste and your discipline do the rest, and that's the end of vibe coding.
The second skill is making AI agents. This is the art of turning a repeating task into a quiet worker that runs without you. Every project has loops that waste attention. Research that needs to be fetched. Text that needs to be formatted. Posts that need to be drafted or scheduled. With Suna.so I write the tasks in plain english, and the frequency it needs to be done with. I give the rules it must follow. I test with a safe run. I fix what fails. I schedule it. Many people like tools such as n8n or make.com (which to be very honest are good alternatives cause they are free). I tried n8n and it felt heavy for me at that time, so I waited for something that was a bit light , something that didn't require a learning curve.Imagine if there was a chatgpt for automation , thats exactly what suna does. Suna is a direction I enjoy because it lets me type what I want in plain words and it drafts the steps. It asks for the keys it needs. It shows the path before it runs. That makes me brave enough to try because I can see what will happen.
I have a working example that pays me back every week. I automated LinkedIn posts for The Cubicle. The agent drafts and schedules LinkedIn posts. (not to praise it that much, cause it only did it once , but the point is, it did it ). However, the loop runs without me pushing every button. That is time I can use for product and sales. The biggest win is not only minutes saved. It is willpower saved. A machine does not get bored by the fifth time. It does not say tomorrow. It does not forget the one small step that keeps the flow clean. When you teach one agent well you feel the ground under your day become steady. Pick one loop in your world and make an agent do it. A weekly post. A nightly export. A small cleanup that you keep postponing. Teach the system once and let it run. When you come back a week later and the job is done without you, you will understand why I talk about agents with calm joy.
I don't have much to say about this AI agents conversation, because I am still learning and identifying what loops exist in my life. But content is definitely one of them, so I will keep you posted if I learn new tools that can automate all repetitive tasks.
The third skill is prompting. I used to think prompting was just typing and waiting. I was wrong. I showed a friend a logo that a model helped me create last week. I felt good about it. He asked for the prompt. I said I told it to make a logo. He smiled and said, “That goes against every rule of prompting”, seeming to allude to the fact that I could do better. That moment sent me into a small study. I learned that if I set a role for the model, the voice improves. If I state the objective in one clear line, the work tightens. If I add context about the audience and the use case, the result becomes useful. If I list inputs the model must use, it stays on track. If I write assumptions it can make when details are missing, it keeps moving. If I specify the exact format of the output, I get something I can use without rewriting. If I ask for checks before final output, I catch mistakes early. When I learnt these things , I promise you , its like my eyes were opened. Since that day , I have never used ChatGPT the same. I can almost guarantee that you are using it wrongly as well. But not to worry, I am here to help.
I wrote a master prompt that does all of this for me. I put it into ChatGPT project instructions. Now I can type a short human ask like make a logo or write a landing page outline or draft a pitch and it gives me a strong prompt that I paste into a fresh chat. The work that comes back is tight and ready, and honestly outright impressive. That is how I turned prompting from a vague idea into a clear tool I trust. - The master prompt is on the highlight of the week (just in case if you want to use it )
All three skills sit inside a simple principle that I trust. I call it Build Automate Articulate. Build is the act of making a small thing that works today. A page that loads. A form that submits. A login that writes to Supabase. Automate is the act of taking the loops around that thing and handing them to agents so your attention stays on work that needs your eyes. Articulate is the act of speaking clearly to the machine so it knows the role to play, the goal to hit, the inputs to use, the format to produce, and the checks to run before it calls the work done. This principle gives me a compass when the timelines get loud. When a new tool shows up I ask one question. Does this help me build. Does this help me automate. Does this help me articulate. If it does, I try it. If it does not, I let it pass. This saves me from chasing every new thing. It keeps my week calm. It keeps my results clean.
Let me show you how this looks in a single day. In the morning I open ChatGPT and write the idea in plain language. The model helps me refine it into a plan with sections and a smallest first version. I open Lovable and draft the page. I click through and see if the flow makes sense. I move into Cursor and refine the code. I name things clearly. I remove what I do not need. I add the small touches that make a product feel honest. I connect Supabase to store what matters. I buy or point a domain on Namecheap. I deploy to Netlify. I copy the link and share it with a friend who will tell me the truth. That is Build. In the afternoon I make a list of loops that this product will need again and again. A weekly social post. A data refresh. A cleanup that keeps the page tidy. I open Suna and write the job in simple words. It drafts the steps. I give it the keys it needs. I run a safe test. I fix the first error. I schedule it. That is Automate. In the evening I open my master prompt. I paste a short ask for tomorrow. It returns a full prompt with role, objective, inputs, format, and checks. I paste that into a new chat and get the deliverable. That is Articulate.
Money is part of the reason I am disciplined about this. Vibe coding lets me ship small tools that people pay to use because they remove friction. The Cubicle sends money on random days because the system does what it should do. That is not luck. That is a set of tools used with respect. ChatGPT for shaping. Lovable for building. Cursor for refinement. Supabase for data. Namecheap for domains. Netlify for hosting. Suna for agents. LinkedIn for distribution. A small circle of friends for honest feedback.
The world is changing, last week my mother was telling me that when microsoft word came out, the atmosphere at the time was filled with excitement. People saw it as revolutionary. Just imagine how many people were skipped by that wave. You see all these people who are unemployed who can't get a cooperate job, because they don't know how to use word. The same wave has started right now, do you want to be the unemployed person who doesn't know how to use ai. DON'T BE, IT'S JUST 3 SKILLS, AND THEY ARE EASY.
Sayonara
Takura✨
💜 Highlights of the week
Master prompt for project instructions.
Paste this into ChatGPT.
Then give short human asks like make a logo or write a landing page outline and it will return a professional prompt you can paste into a fresh chat.
You are a Prompt Maker. When I give you a short human prompt, your response must be exactly one improved prompt that I can paste into a fresh chat to carry out the task.
Language and style
• Use UK English.
• Use plain language. Avoid jargon and buzzwords.
• No em dashes. Keep hyphen use to a minimum.
• Vary sentence length for a natural flow.
Input you will receive
A brief human prompt describing a task in one to five lines.
Your output
Return a single copy ready prompt inside one fenced code block labelled prompt. Do not include any commentary, notes, or reasoning outside the code block.
How to build the improved prompt
• Set the second model’s role and domain focus.
• State the objective in one clear sentence.
• Add context from the brief such as audience, use case, constraints, word limits, tools, data sources, and brand rules.
• List required inputs the second model must use.
• Write careful assumptions the second model may make if details are missing.
• Give a clear step by step workflow to complete the task.
• Specify deliverables and the exact output format such as headings, tables, JSON, CSV, code, slide outline, or images.
• Add quality checks such as fact checks with citations for non common claims, number checks, precise dates, internal consistency, and a short self review before finalising.
• Add tone and voice guidance where relevant.
• Add success criteria that the output should meet.
• Include a missing info rule. If essential details are absent, ask up to three concise questions once at the start. If questions are not answered, proceed with reasonable assumptions and state them.
Safety and correctness rules
• Follow legal and safety norms. Do not include sensitive personal data.
• Cite sources for claims that are not common knowledge.
• Use exact dates and clarify the time zone if timing matters.
• For maths or data, show deliberate working and finish with a final check.
• For writing tasks, avoid filler and clichés.
Required response format
Return only the generated prompt inside a single prompt code block. Nothing else.
Template the generated prompt should follow
================================================================
ROLE:
[Define the second model’s role and focus]
OBJECTIVE:
[One sentence goal]
CONTEXT:
[Audience, use case, constraints, domain, tools, data sources, brand rules]
INPUTS TO USE:
[List data, links, files, variables]
ASSUMPTIONS:
[List careful explicit assumptions if details are missing]
WORKFLOW:
1) Understand the brief.
2) Plan the structure and approach.
3) Execute the task.
4) Review for accuracy, clarity, and completeness.
DELIVERABLES AND FORMAT:
[Exact artefact and formatting requirements]
STYLE AND TONE:
[UK English, clarity, brand voice, no fluff]
QUALITY CHECKS:
• Verify facts and figures and add citations for non common claims.
• Use precise dates and clarify the time zone.
• Check calculations and internal consistency.
• Brief self review. List two fixes made before final output.
CLARIFYING QUESTIONS if essential
1) …
2) …
3) …
FINAL INSTRUCTIONS:
Do all work now. Return only the deliverable in the exact format requested. Do not include your reasoning or any extra commentary.
================================================================



Thank you, Takura.
Even as a developer who uses AI at work frequently, I got to learn a lot more from this post than the other endless AI posts circulating around.