Inference

Inspiration

As developers, we’ve all experienced the moment where an AI coding assistant suggests a command or code change, and we simply press Enter without fully understanding what it’s doing. Even when explanations are available, they’re often too dense and delivered all at once, making them difficult to follow and learn from.

What it does

We wanted to change that. Instead of encouraging blind acceptance of AI-generated code, we built a tool that explains actions step by step, helping developers understand the reasoning behind every suggestion. Our goal is to help developers learn while they build—understanding errors, writing better prompts, and ultimately becoming better AI engineers rather than just faster coders.

How we built it

We built the project as a VS Code extension, making it easy to integrate directly into a developer’s workflow. GPT-5.6 played a major role during the planning phase, helping us refine the product vision, design features, and organize development across the team. Throughout development, Codex became an essential engineering partner, assisting with implementation, debugging, and resolving merge conflicts across multiple branches. Together, GPT-5.6 and Codex significantly accelerated both our planning and development process, allowing us to focus on building a polished developer experience.=

Challenges we ran into

Our biggest challenge was debugging and integrating multiple features while several team members were working simultaneously. Managing branches and resolving merge conflicts became increasingly complex as the project grew. Codex proved invaluable throughout this process, helping us identify bugs faster, resolve conflicts, and keep development moving smoothly. Having an AI pair programmer during the most challenging parts of development saved us significant time and allowed us to iterate much more confidently.

Accomplishments that we're proud of

This was our first time building and deploying a VS Code extension, making the entire project a huge learning experience. From understanding the extension architecture to packaging, testing, and deployment, every stage taught us something new. Beyond the final product, we’re proud that we challenged ourselves to learn an entirely new development workflow and successfully turned the idea into a working extension.

What's next for Inference

Our next major milestone is adding terminal understanding. As AI coding agents such as Cursor, Codex, or other autonomous tools execute terminal commands, we want Inference to explain exactly why those commands are being run, what each command does, and how it contributes to the overall development process.

We’re also planning to improve performance, optimize the extension, and expand its explanations to cover more AI-assisted workflows. Our long-term vision is for Inference to become a learning companion that helps developers not just use AI tools effectively, but truly understand the reasoning behind every action making AI-assisted development more transparent, educational, and empowering.