Using AI to Build an AI App - Part I

The days of "buying software" are ending.

You can build your own CRM or RPG now.

You can use AI to build anything these days.

We’re launching a new series today called Using AI to Build an AI App to showcase how the barriers to building tools, systems and entirely new technology are falling fast.

I believe the fate of software companies is to become AI builders.

I don’t mean using AI to write code and maintain databases… that’s happening already. I mean all software companies will STOP writing software… and focus 100% on building and enhancing the AI agents and platforms they control.

“but Matt, isn’t this semantics… aren’t AI agents just collections of Python code?”

Indeed, this is true… but the companies won’t be focused on creating deterministic programs for their users to run. They will be focused on building:

• User interface agents

• User experience agents (bosses of the UI agents)

• Routing agents

• Control flow / logic agents (bosses of the Routing agents)

• Database operations agents

• Product utilization monitoring agents

• … and likely dozens of other flavors of agent focused on things like user engagement, user net promoter score and more

You see these agents will be the the ones writing the software.

Dynamically, for each user across every interaction.

AI will be materializing UI and driving incredible UX while handling all the control flow logic and backend operations... in real-time.

Isn’t That A Bad Thing for Devs?

It is true that many of the massive tech companies with thousands of headcount will choose to take advantage of the speed and scale AI provides and “gain agility” over the coming years by slowing hiring, selling off non-core divisions and establishing lots of small pods with AI being used to collaborate and gain efficiencies between them.

This means more firms will be born.

Tight teams of 3 to 5 builders and passionate communicators will be able to compete for attention (and clients) with the Tech Giants.

These builders will have armies of AI agents at their fingertips ready to work (24/7) ideating, designing, developing, engineering, implementing, testing, refining, rebuilding and never stopping until all user requirements are satisfied.

many willing helpers

This will expand the surface area for development and invite a more diverse chorus of voices into the global feedback loop of engineering that’s accelerating all around us.

This means more opportunities for developers, not less. In this future world you have 20,000 viable companies to explore and potentially build for as opposed to the current crop of 20 A+ companies, about 50 As, 100ish Bs and maybe 250 Cs.

That’s a 50x increase in opportunities worth exploring as a developer.

Cutting Edge Development

So are AI agents running around laying down software railroad tracks just as the user gets there today?

No, not yet.

We do have development platforms in the market that use agents to write code, write unit tests, document codebases and more… but they aren’t collaborating in swarms without human intervention on a mission to “build and constantly upgrade the car around the driver as they lap the racetrack”.

Google have released IDX, an integrated development environment that features Gemini AI as partner, I did a dive into this platform back in May:

IDX is attempting to integrate most of the features you’d expect from an IDE, but extend the power of AI and the richness of Google Cloud Platform to make deploying your recently built applications more intuitive.

Just yesterday I gained access to Gemini’s answer to OpenAI’s GPTs, which Google has branded Gems.

These are LLM instances that have very specific system prompts (you are a world-class coder and you are coding as a partner with the User, you will help them plan, design, develop, engineer and implement code, etc…) that drive experiences that are tailored to different scenarios.

A creative writing partner.

A software engineering partner.

A partner for every project, passion and mission.

I passed this prompt to my coding partner gem:

Build a Next.js web app.

The app is called Wealth Copilot. It uses generative AI to help the user plan their wealth building and manage it too. There are three parts. There's a chat UI for the AI conversation. There's also a visual showing the status of each wealth engine (dividend portfolio, e-commerce business, real estate, consulting, etc..) and the last part is called Wealth System Summary and it shows net worth, income growth rates, net worth growth rates, expense growth rates.

When I ran this prompt I was pleased to receive back not a quick, dirty (and very half-baked) series of scripts for the three pieces I mentioned… instead my AI coding partner wanted to plan this out by talking about it first.

Measuring and even optimizing BEFORE coding… just like real engineers do.

My new best friend starts by reading back the idea at a high-level to me.

Perfect.

Then we discuss the broad strokes it will take to deliver this vision.

The system even explains calls out key considerations I need to consider before ending with a call-to-engagement to continue the iterative process.

Brilliant.

I also noticed that I forgot to tell Gemini that I was using Project IDX. Hopefully this accelerates my development process. I also want to make the database schema more clear for my AI coding partner. Here’s what I said next:

I am going to develop this using Google Project IDX.

Going to use Gemini API for the AI.

Let's use a database to store the financial information. I suggest a table to store each wealth engine's data, then a separate table for the wealth system summary data.

Ultimately I want to deploy this application on Google Cloud.

My buddy replied positively. AI systems almost always do. In fact you need to program your systems to give you critical feedback: “tell me 5 things I didn’t do well with this code, and suggest 3 improvements for each”. I’ll ask it to yell at me and my code later though, once I have some.

During the ideation and design phase the positive energy is a major plus.

It calls out potential table structures as I immediately realize we’ll need a third table for user data. Thankfully I can update this in seconds with AI and zero sighs or “why didn’t you think of this sooner?” which humans often specialize in.

The output ended like this — with a satellite view of the development roadmap:

“Matt, why are you building software if the days of selling software are over?”

I plan on using this personally and sharing it with friends which includes anyone who’s trying to organize and optimize their personal generation and storage of wealth.

The Road Ahead

In Part II we’re going to lay out the actual plan to build this copilot application.

Soon we’ll turn Wealth Copilot by WealthSystems into an intuitive and effective AI-powered wealth management and wealth building platform.

🥺 Please follow our newly launched video channel Hacking Wealth, we need to reach a 100 followers before Rumble will unlock a named URL.

Thank you for helping us grow and thank you for reading Life in the Singularity.

I started this in May 2023 to track all the accelerating changes in AI. New players, new models, new methods… a big new future materializing in front of us.

Our audience now includes Big Tech engineers and executives, incredible technologists, Fortune 500 board members and thousands of people looking to leverage technology to maximize the utility in their lives.

To help us continue our growth, would you please Like, Comment and Share this?

Thank you again!!!