first project done✅ -> AGIpath(.net)

Product development has dramatically changed recently due to advancements in AI.

Innovations like Cursor AI, Bolt.new, and general AI breakthroughs—especially those demonstrated with o3 (we’ll see below)—are becoming the driving force behind our progress.

Let’s use the start of 2025 to become an unstoppable force of nature, together.

Before diving into why o3 is almost AGI (I’m convinced we’ll reach AGI this year), I’d like to share how, by January 3rd, I’ve already developed my first project: AGIpath.net. (Give it a click, but come back 😉 )

I’ll walk you through my 7-step project blueprint—a systematic approach to developing projects that improve and evolve over time.

My 7-step blueprint for building projects

I used this to build AGIpath.net at warp speed.

1. Project start with Bolt.new

Bolt.new by StackBlitz, enables AI-powered full-stack web development entirely in-browser, with tools for scaffolding, code generation, and even deployment.

I only use it to create a foundational version of what I plan to build. You completely circumvent the blank-page problem of not knowing where to start. I simply describe what I want to create and maybe how I want it to look, along with any other relevant information. After a brief moment, it provides a solid first version of the project.

Fantastic: Bolt.new is now collaborating with Supabase—a great AI-driven database (it builds a PostgreSQL DB). With one click, it can be hooked into your webpage.

2. Iteration with Cursor AI

You can easily download the complete codebase from Bolt.new. That’s what I do, and I reopen it with Cursor AI—my absolute favorite AI product of 2024! It’s an IDE that integrates AI in the most effective way to develop and refine the code you’ve just downloaded.

Now we build the page: adding buttons, workflows, text, images, and anything else you need.

Tip: Make sure to link your other existing pages—this network effect will help rank them higher!

[I wrote about Bolt.new and Cursor AI in earlier episodes: one, and two.]

3. Make the project available to the whole world

A professional product needs its domain. I bought AGIpath.net via Squarespace.

Next, if you want to deploy your developed code on a hosting service—which can get quite complex—I maintain a conversation with Cursor’s AI Agent to guide me.

• I tried DigitalOcean’s Spaces Object Storage (€5/month), but you can’t connect your domain to it. The next option costs €20/month.

• Being resource-conscious, I tested Netlify and GitHub Pages in parallel. GitHub Pages was an absolute charm to deploy. Connecting the domain name took just 5 minutes. Best of all, it’s practically free—all you need is a GitHub premium account for €3.80/month, and you can host as many pages as you want. 🤯

4. [Repeatable task block] Iterate the project with Cursor

This step is necessary because you’ll identify issues to work on through monitoring (next step), error messages (e.g., in the browser), your own observations on web and mobile, and feedback iterations with trusted people.

5. Featurebase

I appreciate the work of Pieter Levels and found that he uses Featurebase for all of his projects. It’s a simple concept and mega helpful: your users can suggest features or report bugs as they use your page, including upvoting.

This is a super powerful source of information for you. When improving the page, it should be your first reference.

6. Monitoring

Register your app on these platforms:

• Ahrefs: Optimizes SEO by analyzing backlinks, keywords, and competitors.

• Google Analytics (GA): Tracks and analyzes website traffic and user behavior.

• Google Search Console (GSC): Monitors site health, indexing, and search visibility.

If you have good stats in Ahrefs and GSC, you’ll observe an increase in traffic in GA!

Tip: GA has a great mobile app. This is what it looks like for me after the first few days:

7. Launch! (You can launch multiple times)

When your project is good (not perfect), you launch. (I haven’t done it yet for AGIpath.net, but I’ll probably do it now with this email.)

You launch on Reddit, Hacker News, your newsletter, X.com, LinkedIn (though no one cares about it anymore), and other platforms. You probably can’t launch everywhere—there’s just too much—but you must launch where your potential users are. Your job is to figure that out.

For me, X.com is becoming more critical.

Before you launch, though, make sure your page can handle a lot of traffic for a while and won’t crash. How? Ask Cursor AI how to upscale your resources.

Tip: You can launch multiple times. New feature implemented? Re-launch!

If you continuously ship app improvements, you’ll be unstoppable!

In 2025, everything changes!

OpenAI’s o3 is close to AGI

o3 is the succeeding model of o1 (they skipped o2, because of potential confusions with o2 the company).

It uses what OpenAI calls a "private chain of thought," where it pauses to internally deliberate before generating responses, aiming to simulate human reasoning processes more effectively.

It is solving the ARC-AGI benchmark so good, that they have to think of new problems to solve.

What is the 𝗔𝗥𝗖 𝗔𝗚𝗜 𝗦𝗰𝗼𝗿𝗲?

The Abstract Reasoning Corpus (ARC) tests AI’s human-like intelligence by evaluating adaptability and generalization on puzzles that can’t be solved by brute force or overfitting. Created by François Chollet, ARC measures how well AI handles novel tasks.

2024 (Early): GPT-4o →Score: 5%

☝️GPT-4o made little progress on the ARC-AGI benchmark, showing that making models bigger isn’t enough to solve these puzzles.

2024 (Later): o1 and Hybrid Methods → Score: 32%

• Hybrid Approach: Jeremy Berman combined Claude 3.5 Sonnet, genetic algorithms, and a code interpreter to reach 53%, the highest score before o3.

☝️These methods improved performance by blending AI language models with algorithmic tools, achieving meaningful, incremental progress.

2024 (A Few Days Ago): OpenAI’s o3

• Standard Score: 75.7%

• High-Compute Score: 87.5%

☝️o3 made a massive leap in ARC-AGI performance through program synthesis and advanced reasoning, far surpassing earlier models. This represents a significant architectural breakthrough.

Not AGI—Yet

While o3 is impressive, it’s not AGI.

It relies on external verifiers during inference. → solved quickly

It lacks autonomous learning, a key trait of human intelligence. → for this we need an architectural adjustment.

☝️o3 is a big milestone, but there’s still some way to go before achieving true AGI.

Chollet and team (from arc prize), are working on a second version of the benchmark, coming soon. [Highly recommend reading this article if you want to know more.]

2 more thoughts:

1. Scale of systems like o3 comes faster than anyone can imagine. (👀 DeepSeek v3.)

2. Given that there is no wall in the pretraining of AI (proof point o3), and the massive investments in the field, I am convinced we’ll see v0 of AGI this year.

Happy New Year,
Martin 🙇

• I recommend:

  • Beehiiv if you write newsletters.

  • Superhuman if you write a lot of emails.

  • Cursor if you code a lot.

  • Bolt.new for full-stack development.

  • My book - Generative AI: Navigating the Course to AGI.

  • Follow me on X.com.

• AI for your org: We build custom AI solutions for half the market price and time (building with AI Agents). Contact us to know more.

You might like our last episodes: