Passer au contenu principal

Code to Content - From Pages to Systems

Brandon Qin’s follow-up demo extends Code to Content into a fully AI-generated website MVP, using shared inputs, memory, and feedback loops to autonomously maintain structure, content quality, and consistency across abnormal.ai.

NOTE: Demo visuals include blurred data or synthetic placeholders to protect customer privacy.

Why Pages Weren't Enough

Brandon starts with a simple scenario.

A feature ships. A PMM is responsible for getting it onto the website.

The work begins in GitHub, reading through a PR to understand what changed. Then moves to Jira for additional context and customer stories. From there, the PMM must determine where the feature fits within the website's structure and how to position it. They still need to:

  • Follow internal marketing guidelines
  • Navigate the CMS to structure the page correctly
  • Coordinate in Slack with engineers and product teams
  • Schedule time to clarify gaps

Code to Content EP3 Brandon Qin Screengrab1 TC0 27

It is not one task. It is a chain of dependencies.

Three frictions show up consistently:

  1. Understanding a feature requires stitching together multiple fragmented sources
  2. Content creation involves multiple rounds of drafting, review, and alignment
  3. Website pages fall out of date as features ship faster than content can be produced

To quantify the problem, Brandon asked Nora to analyze how long it actually takes for code to become customer-facing content. The answer was consistent across workflows: it typically takes two to three weeks from merge to publish.

When you break that down, it adds up to roughly 432 hours across the pipeline, including drafting, reviews, approvals, and coordination. That is the system this work is replacing.

A New Architecture

Code to Feature Pages introduces an end-to-end pipeline that starts at a merged PR and ends with live website content.

Code to Content EP3 Brandon Qin Screengrab2 TC0 39

The flow is structured:

  • Ingest: Capture merged GitHub PRs along with linked Jira tickets and contextual signals
  • Classify: Map changes to specific product areas and sub-features
  • Readiness check: Ensure only production-ready, customer-facing features move forward
  • Group: Combine related PRs into a cohesive customer story
  • Generate: Use LLM reasoning to create structured marketing content
  • Publish: Push content into the CMS and update the website

Each step reduces ambiguity. Instead of manually interpreting changes, the system builds a consistent path from code to content.

Brandon’s framing is direct: if a PR merges, the website should reflect it within minutes.

What the Demo Shows

The demo highlights a key limitation of the current site.

Inbound email security, one of Abnormal’s core products, is represented on a single page. That page attempts to capture a wide range of features, making it difficult to fully communicate what the product does.

The MVP restructures this approach.

Code to Content EP3 Brandon Qin Screengrab3 TC0 48

Instead of a single page, the website is broken into granular sub-feature pages, each aligned to specific parts of the codebase.

In the prototype:

  • Product areas expand into sub-features
  • Each sub-feature is represented as its own page
  • Pages are generated directly from real product changes

When a new PR merges, the system updates the relevant page. It does not simply append new information. It rewrites the content to incorporate updates coherently.

This creates a continuously maintained representation of the product.

Scaling Autonomy with Guardrails

Brandon was clear that autonomy does not remove humans from the loop. It changes where they spend time. The next phase focuses on tightening evaluation and governance so quality scales alongside automation.

Upcoming work includes:

  • Human review gates for content quality
  • Shared gold standards for layout and structure
  • Improved sitemap hierarchy with PMM input
  • SEO, accessibility, and legal considerations
  • Continuous iteration toward full autonomy

The ambition is explicit: a website that stays current by default, improves with use, and reflects what Abnormal ships in near-real-time.

One System, Many Surfaces

Code to Feature Pages creates a direct connection between what is built and what customers actually see. Instead of relying on manual translation across tools, teams, and handoffs, the system treats code as the source of truth and systematically turns it into customer-facing content. As a result, every merged PR has a clear path to becoming a website update, reducing lag and ensuring the site reflects the product in near real time.

Keep exploring

Browse more workflows or follow other series.