Think Outside the Bots

Inspiration

With the advent of ChatGPT and large language models, we've entered a new era. The way we learn, solve problems, and create software has fundamentally changed.

But one thing caught my attention.

In our pursuit of convenience, we've started removing friction from learning. AI gives us answers instantly, and while that feels like progress, not all friction is bad. The space between knowing the answer and arriving at the answer is where real learning happens.

Think about mathematics. We ask students to show their work because we care about the reasoning—not just the final answer. That process builds intuition, confidence, and the ability to think independently. Today, many learners jump straight to AI-generated solutions and miss the journey that develops those skills.

This became the motivation behind Think Outside the Bots—an attempt to make AI a thinking partner instead of an answer machine.

What it does

Think Outside the Bots introduces intentional cognitive friction into AI-assisted learning.

Instead of immediately providing answers, the platform guides learners through structured thinking workflows before revealing the solution. These workflows encourage learners to actively engage with a problem, reflect on their understanding, and build conviction in their reasoning.

Current workflows include:

Feynman Technique – Explain the concept in your own words before receiving AI feedback.
Socratic Questioning – AI challenges assumptions through guided questioning.
Draft First – Attempt your own solution before AI reviews it.
AI Freeze – Introduce a configurable pause before AI responds, encouraging learners to think first.
Spaced Repetition – Reinforce concepts over time through active recall.

The platform also supports:

Single or combined learning workflows
AI-adaptive workflow selection
Bring Your Own LLM support
A visual workflow builder for educators to create and customize their own learning experiences

The goal is simple:

Think with AI, not through it.

How we built it

The prototype was built over a single weekend using Spec-Driven Development with GitHub Speckit, Codex, and GPT-5.6.

Rather than using AI purely to generate code, I used it as a collaborative design partner. GPT-5.6 helped refine the problem statement, iterate on product ideas, challenge assumptions, and translate specifications into working features. Spec-driven development provided a structured way to rapidly evolve the application while keeping the implementation aligned with the original vision.

Coming from a Python and data engineering background, this project also became an opportunity to quickly learn the modern React ecosystem and frontend tooling.

Challenges we ran into

The biggest challenge wasn't writing code—it was designing the right learning experience.

Finding the right balance of cognitive friction took several iterations. Too much friction makes learning frustrating, while too little makes the experience no different from a traditional AI chatbot.

Another challenge was understanding the rapidly evolving GPT-5.6 model family. Different models have different capabilities, token limits, and pricing characteristics, so learning when to use each model and optimizing prompt flows around their token usage took time during development.

Since my background is primarily in Python and data engineering, I also had to quickly get up to speed with React, pnpm, component-driven development, and workflow state management.

Accomplishments that we're proud of

Built a functional prototype over a single weekend.
Created a configurable workflow engine that supports multiple learning strategies.
Designed a visual workflow builder that allows educators to build and customize their own learning experiences.
Added Bring Your Own LLM support so learners can use their preferred AI models.
Demonstrated how AI can encourage reasoning before providing answers instead of replacing the thinking process.

What we learned

This project reinforced that AI is most valuable when it augments human thinking rather than replacing it.

I also learned how effective Spec-Driven Development can be when combined with modern AI coding tools like Codex. More importantly, GPT-5.6 proved to be far more than a coding assistant—it became a genuine thought partner that helped shape the product itself.

What's next for Think Outside the Bots

This prototype is just the beginning.

Future plans include richer adaptive workflows, mind maps, multimodal interactions, collaborative workflow sharing, educator analytics, and deeper personalization based on learner progress.

As AI becomes better at giving answers, I believe we should become even better at asking questions and thinking for ourselves.

Think Outside the Bots is a small step toward that future.