3 Years of AI Mistakes Taught Me One Thing

I've been completely obsessed with AI since ChatGPT 3.5 came out in late 2022. We could finally have a conversation with an AI that knew the entire internet!

Wow!

Back then, just that alone seemed incredible... simply having a conversation.

So naturally, I tested every new model as soon as it dropped:

• ChatGPT 4 by OpenAI in March 2023
• Then Claude 2 in July 2023, Anthropic's first consumer version

The race for new generative AI models had truly begun.

We kept seeing increasingly powerful AI models roll out:

• Reasoning (to improve response quality)
• Multi-modality (to understand text, images, and video)
• Larger context windows for messages sent (and received)
• More memory to remember conversations
• Tools attached to models to perform specific actions
• Models running locally on computers without internet
• AI agents working autonomously!

And that was just the beginning.

Being the geek I am, I tried every tool the moment it launched.

I tested every model.

I built automations on top of automations.

And I made every mistake you can make.

Explaining the same thing 200 times

You know that feeling when you're explaining your business to ChatGPT for the fifteenth time?

"I have a company that... My target audience is..."

Yeah, we've all been there.

Every conversation started from scratch.

Every new session forgot everything we'd discussed before.

So I built my own text database to copy-paste my context. I crafted elaborate prompt templates. I created "memory" documents I'd paste into every conversation.

It worked.

Sort of.

Until it didn't. Too much context, too many prompts, too much copy-pasting...

Smart automations?

So I started building automations to set up context and tools once and for all.

I created Zapier workflows connected to Airtable databases connected to Notion pages connected to... even more Zapier workflows.

I built systems so complex I couldn't remember what they did.

I had automations triggering automations triggering automations.

It felt productive.

But I had tons of errors and it wasn't much more practical.

After all that, I realized I only needed a collection of text files (markdown for the techies).

The simplest solution (I should have thought of it sooner).

The vibe-coding trap

Then I fell into another trap where I wasted even more time.

I'm a perfectionist by nature.

And when you can build things, you want to build the perfect thing.

So I spent an entire year vibe-coding to build my AI app that would let me manage everything: context, automations, AI agents, you name it.

An ENTIRE year.

I was building a platform to create custom agents with different personalities. Knowledge bases that worked like communities. Integrations with everything. Features nobody had asked for.

It was going to be amazing.

It was also going to take forever.

I kept adding. Kept perfecting. Kept telling myself "just one more feature."

The tool was never ready because I kept making it more complex.

Then something clicked: I don't need a complex platform. Nobody does.

What I actually need—what every entrepreneur actually needs—is dead simple:

Context + Automation. In text format (markdown).

That's it. Nothing more, nothing less.

The key was to build a solid foundation first. Everything else would follow: integrations, capabilities, content repurposing across different formats.

All of that builds on a simple, flexible, and most importantly private tool to manage your context and systems.

A ridiculously simple solution

After 3 years of all this, you know what I use now?

Text files.

That's it.

Markdown files with my context. My voice. My preferences. My goals.

Obsidian for local notes.

And AskMojo (the app I'm building) for AI context.

No complex automations. No database schemas. No workflow builders.

Just text that AI can actually understand.

To adapt to AI's constant evolution, we need simplicity and flexibility

Every time I switched tools or rebuilt my system, I lost context. Every "upgrade" meant starting over.

The entrepreneurs getting real results with AI aren't the ones with the fanciest setups. They're the ones who've built context that compounds over time.

Your AI should know more about you and your business six months from now than it does today. Not because the model improved—because your context got richer.

Why I'm building AskMojo

When Anthropic released MCP for tool integrations and Skills for workflows, something clicked.

The infrastructure is finally here to build what I've always wanted: a system that remembers and gets better the more you use it.

Not another AI tool you'll replace in six months. A foundation for everything that comes next.

I'm building AskMojo for entrepreneurs who are tired of starting over. Who want their AI to truly understand their business. Who care about long-term context—not just whatever model dropped this week.

Simple. Flexible. Private.

Context and automation in markdown (in text). That's the whole philosophy.

What comes next

Once the foundation is solid, everything becomes possible:

• Video clones that talk like you.
• Voice agents that handle calls.
• AI community managers that engage your audience while you sleep.
• Email assistants that draft responses in your style.

All of that is coming. But none of it works without solid context to rely on.

That's the lesson I learned the hard way.

You can't build a video clone without teaching it who you are.

You can't deploy a voice agent without giving it your knowledge.

You can't automate community management without defining your style and goals.

Context is the foundation. Everything else builds on top.

AskMojo starts with context and automation. The rest follows.

The beta launches in January.

If that resonates with you, join the waitlist.