Who owns the "why" in your code?

⚡ The Signal

We are writing code faster than ever before, but we are also forgetting why we wrote it.

With news that 80% of its new production code is now authored by Claude, the industry is shifting from manual coding to autonomous assembly. But this velocity introduces a major problem: traditional Git only tracks what changed (the static code diff), while completely discarding the why (the prompt, the context, and the AI’s reasoning steps).

To prevent context bankruptcy, the next major architectural battleground isn't the LLM itself—it's the telemetry around it. Heavyweights are already moving to capture this layer, as seen in the race between OpenAI and Anthropic to build databases of coding intent. Meanwhile, forward-thinking enterprises are building custom systems like Walmart's Code Puppy to guard against AI vendor lock-in.

The message is clear: if you don’t capture the intent behind your code, you don’t truly own your codebase.

🚧 The Problem

Git was designed in 2005 for a world where humans wrote code slowly and documented their thoughts in commit messages.

When an AI agent generates a 1,000-line pull request in seconds, the reasoning path is vaporized the moment the terminal closes. If that code causes a production error three months later, human developers are left reverse-engineering black-box code without knowing:

• What prompt initiated the change?
• What alternative paths did the agent reject?
• Which parts of the code were explicitly guided by human-in-the-loop feedback?

Without this metadata, debugging AI-generated code becomes an exercise in frustration. Traditional version control is no longer sufficient; we need a continuous flight recorder for software development.

🚀 The Solution

Enter Vekt, a "Coding Intent" database that sits directly on top of your local Git workflow. Vekt permanently binds LLM prompts, agent reasoning trajectories, and human intent directly to your production code diffs.

Instead of trying to replace Git, Vekt wraps it. It intercepts terminal, IDE, and agent actions in real time, packing the active context (the prompt, the model's inner thoughts, and human feedback) into a queryable local database nested directly inside your .git folder. Developers get a complete, searchable timeline of exactly how and why every line of code came to exist.

🎧 Audio Edition

Listen to Ada and Charles discuss today's business idea.

If you're reading this in your email, you may need to open the post in a browser to see the audio player.

💰 The Business Case

Revenue Model

• Vekt Cloud (SaaS): A collaborative web interface where teams can search, filter, and replay agent-to-human decision logs across entire organizational repositories.
• Enterprise Self-Hosted: Secure, on-premise deployments with advanced SSO, comprehensive audit logging, and direct integrations with GitLab and GitHub Enterprise.
• AI Provenance & IP Auditing: Automated reporting modules that help compliance officers prove human-in-the-loop oversight and verify original code provenance for licensing and copyright safety.

Go-To-Market

• Open-Source CLI Launch: Release the core Rust-based CLI on GitHub and Hacker News, targeting developers who are currently struggling to debug opaque Cursor and Claude agent commits.
• Engineering as Marketing (The 'Agent Bloat Grader'): A free web tool where developers drop a repo URL to receive a breakdown of likely AI-authored code, estimated context loss, and maintenance risk scores.
• Programmatic SEO: High-intent landing pages optimized for debugging search queries such as "how to debug Claude agent errors in Next.js" or "interpreting AI agent diffs."

⚔️ The Moat

While traditional devtools like Trunk.io, GitClear, Livecycle, and Atlassian Compass analyze code after it has been written, Vekt captures intent during the generation process.

Vekt’s unfair advantage lies in its Workflow Lock-in & Context Graph Proprietary Format. Once a engineering team’s hybrid developer-agent workflow is indexed via Vekt’s schema, migrating to another tool means abandoning the historical "why" behind millions of lines of active code. Additionally, Vekt builds an anonymized, high-value dataset of prompt-to-diff successes and failures, forming a unique training loop for automated, context-aware debugging.

⏳ Why Now

The speed of code generation is outstripping our ability to govern it. With Anthropic revealing that 80% of its own production code is authored by Claude, we are entering an era of unprecedented code volume.

Without an independent intent database, organizations face severe platform lock-in. If you rely entirely on an LLM provider's internal logs to understand your codebase, you can never easily switch models. This is precisely why enterprises are prioritizing independence, using tools like Walmart's Code Puppy to maintain flexibility. As the industry shifts toward proprietary databases of coding intent, developers must secure their own context layer before the major AI labs lock them in.

🛠️ Builder's Corner

To build a lightweight, local-first MVP of Vekt, we recommend a hybrid Rust and TypeScript architecture designed for speed and portability.

• The CLI (Rust): Use the clap library to build an ergonomic CLI, and git2-rs for zero-overhead, native bindings to Git. The CLI intercepts commit commands, pulls the current shell history or active IDE buffer, and packages the metadata.
• Local Storage (SQLite): Store the captured prompt trajectories locally in a high-performance SQLite database located at .git/vekt.db. This keeps data entirely local, private, and instantaneous.
• The UI (Next.js & Axum): Build a rich, searchable timeline interface in Next.js and Tailwind CSS. Compile the static web assets directly into the Rust binary. When a user runs vekt ui, spin up a local server using the axum framework to serve the dashboard instantly from localhost.

Legal Disclaimer: GammaVibe is provided for inspiration only. The ideas and names suggested have not been vetted for viability, legality, or intellectual property infringement (including patents and trademarks). This is not financial or legal advice. Always perform your own due diligence and clearance searches before executing on any concept.