Over the past few weeks, I have somehow found myself building my own version of FRIDAY.
Yes, that FRIDAY.
The AI assistant from Iron Man (huge Marvel fan here).
Unfortunately, mine cannot currently operate an Iron Man suit, run facial recognition across an entire city, or launch satellites.
But she can tell me how much revenue I have closed, remind me what I am supposed to be working on, help me keep track of my finances, send me a daily briefing and slowly become the central operating system for my life.
And that is actually much closer to what I wanted.
It Started With a Very Simple Problem
I have information everywhere.
My work lives across emails, notes, spreadsheets and conversations. I created systems in Notion, Drive, and my company even has our own local hard drive system to store our documents.
My finances are spread across bank accounts, credit cards, investments, CPF and various dashboards.
My personal projects have their own systems.
My health goals are somewhere else.
Then there are all the random things sitting inside my brain:
Remember to follow up with that school.
How much revenue have I closed this year?
How much money did I spend this month?
What am I supposed to do next?
I am actually pretty good at generating ideas, starting projects and seeing possibilities.
What I am considerably less excited about is maintaining fifteen different systems just to remember what all those ideas were.
So I started wondering: what if I had one place that knew what was going on?
Not another productivity app.
Not another dashboard I would abandon after three weeks.
An assistant.
Something I could simply talk to. Could I have my own J.A.R.V.I.S. (now FRIDAY) like Tony Stark?
The First Version Was Basically a Work Assistant
Initially, FRIDAY was mostly about my job.
I work in business development, which means there are constantly opportunities moving at different speeds.
Some clients are ready to close. Some need proposals. Some need follow-ups. Some disappear for three weeks and suddenly reappear asking if we can run a programme next Tuesday.
At the same time, I have an annual sales target to hit. So I wanted FRIDAY to understand things like:
- how much revenue I had already closed;
- how far I was from my target;
- which opportunities were still in my pipeline;
- what proposals were overdue;
- which clients needed follow-ups;
- and, most importantly, what I should focus on next.
Eventually, I built a daily brief. Instead of opening five different places every morning and trying to reconstruct my priorities, I wanted FRIDAY to tell me something like:
Here is where you are. Here is what needs attention. Here is the most useful thing you can do next.
That small shift changed how I thought about the entire project. Because FRIDAY was no longer just answering questions. She was helping me orient myself.
Then the Project Got Slightly Out of Hand
Naturally, once I realised this could work for my job, I started thinking: why stop there?
I wanted my financial life inside FRIDAY too. I wanted to see my salary, commission, savings, investments, CPF, assets, liabilities, debts and spending. I wanted to know my net worth. I wanted to see how many months it would take to clear a debt. I wanted to know whether I was actually getting closer to the financial life I kept saying I wanted.
Then came health. Could I have something that understood my calorie intake, weight-loss target, exercise, hunger, sleep and menstrual cycle instead of treating every day as if my body operated exactly the same way?
Then content. Then personal projects. Then reminders. Then Telegram. Then WhatsApp. Then voice.
At some point I realised that I was no longer building a chatbot. I was trying to build a personal operating system.
"Couldn't Claude Just Do All of This?"
This was the question that almost killed the project.
Because after spending hours setting FRIDAY up, I realised something uncomfortable. Claude could do a lot of it. ChatGPT could do a lot of it too. Modern AI tools can read documents, connect to services, analyse information, write reports and increasingly perform actions on your behalf.
So why was I going through all this trouble? Why build FRIDAY at all?
I eventually realised that I had been asking the wrong question.
The question was never "Can Claude do this?" — it was "Who owns the system?"
Claude is an incredibly capable intelligence layer. So is ChatGPT. But FRIDAY is not supposed to compete with them. FRIDAY is the layer that connects everything around them.
My data. My workflows. My dashboards. My rules. My preferred AI models. My communication channels. My automations. My context.
Today I might use Claude for one task. Tomorrow another model may be better. FRIDAY gives me the possibility of swapping the intelligence while keeping the system.
That distinction suddenly made the hours of setup make much more sense.
The Real Value Is Context
One thing I have learned from using AI extensively is that intelligence without context has limitations.
I can ask an AI: what should I work on today? But unless it knows my projects, deadlines, sales pipeline and goals, it can only give me generic productivity advice.
The question becomes much more powerful when the system already knows:
- I have a S$600,000 sales target.
- This is how much I have closed.
- These proposals are outstanding.
- This client has not replied.
- This project is overdue.
- This opportunity is worth S$20,000.
Now "what should I work on today?" becomes a completely different question. The answer can actually be specific.
That is what I want FRIDAY to become. Not an AI that knows everything about the world — there are already plenty of those. I want an AI that increasingly understands my world.
I Also Learned That Building With AI Is Weirdly Accessible Now
I am not a software engineer.
A few years ago, the idea of building a personal AI assistant connected to dashboards, APIs, messaging platforms and databases would have sounded completely ridiculous to me.
Now I can describe what I want in normal language. Then AI helps me figure out the architecture. It writes code. I test it. Something breaks. I send a screenshot. We fix it. Something else breaks. We fix that too.
Half the time, I am learning what something like an API token or environment variable is approximately thirty seconds before I need to use it.
It is a strange way to build software. But it works.
And I think that might be one of the more interesting changes happening with AI. The barrier between having an idea and building a working version of the idea is collapsing. You do not necessarily have to become a programmer before you start creating software. You can learn while building.
The Hardest Part Isn't Actually the AI
Ironically, the AI itself might be the easiest part. The difficult questions are things like:
What information should FRIDAY remember? What should she be allowed to do automatically? What should require my approval?
Which information should live in Notion? Which information belongs in a database? What should appear on a dashboard?
When should FRIDAY proactively message me? When should she stay quiet? Which AI model should handle which task? What happens when an integration breaks?
Those are not really artificial intelligence questions. They are system design questions. And building FRIDAY has forced me to think about how I actually operate.
The Goal Isn't Maximum Automation
There is a tempting version of this project where FRIDAY does everything. Replies to every email. Schedules everything. Makes decisions. Sends messages. Moves money. Runs my entire life while I lie dramatically on a sofa somewhere.
I don't think that is actually what I want.
I want something closer to augmented decision-making. FRIDAY should notice things. Organise information. Surface opportunities. Prepare decisions. Recommend actions. Handle repetitive administrative work.
But there are areas where I still want the final decision to belong to me.
The goal is not to remove myself from my life. It is to remove enough cognitive clutter that I have more capacity for the parts of my life that actually require me.
FRIDAY Is Still Very Much a Work in Progress
She breaks. A lot.
Sometimes Telegram stops responding. Sometimes a session resets. Sometimes an API behaves differently from what I expected. Sometimes I realise the elaborate feature I spent an hour setting up could have been done with an existing product in five minutes.
And occasionally I reach the deeply philosophical stage of software development known as:
"Why is it still typing?"
But every time something works, the vision becomes a little clearer. I can already imagine waking up and receiving a briefing:
Here is your schedule. Here are the three things that matter today. Here is your current revenue. Here are the clients you should follow up with. You spent more than usual yesterday. Your investment portfolio moved significantly overnight. You slept poorly, so perhaps today is not the day to schedule five difficult conversations. Your dinner ingredients are already at home. And by the way, you have a bill due tomorrow.
That is the FRIDAY I am trying to build.
Maybe Personal AI Won't Be One App
One of the biggest assumptions I had when I started was that eventually there would be one perfect AI application.
Now I am not so sure. I think the future might look more like layers. One model for reasoning. Another for coding. Different services holding different types of data. APIs connecting everything together. Automations running quietly in the background.
And somewhere above all of that, a personal interface that understands what matters to you.
For me, that interface is called FRIDAY. Someone else's might have another name entirely.
What Building FRIDAY Has Really Taught Me
I started this project because I wanted an AI assistant. What I accidentally started learning was how systems are built. How APIs work. How AI models differ. How databases store context. How automation changes workflows. How software products are designed. And, strangely, how my own brain works.
Because before you can tell an AI how to organise your life, you have to decide how you want your life organised in the first place.
FRIDAY is nowhere near finished. There is a very long list of things I still want to build. But I think that might be part of the fun.
For the first time, I am not waiting for someone to release the perfect productivity system for me. I am building my own.
And every few days, it gets slightly closer to the version of FRIDAY I imagined when I started.
Minus the Iron Man suit.
For now.