Websites That Adapt to Reality, Not Memory

A few days ago on X I found this article from Erik, it spoke about how to make a website that rebuilds itself. The idea was interesting and I wanted to try something similar as well.

Erik's approach is honestly brilliant. Every day at 6 AM UTC, his site kicks off a GitHub Action that fetches fresh data from GitHub, Spotify, X, and even the weather. Then it runs four different AI models in parallel - Claude Opus, Composer, GPT Codex, and Gemini - each creating their own version of the site. The outputs get sandboxed to a generated/ folder, archived to public/builds/, and deployed automatically. The site is live before he wakes up.

What caught my attention wasn't just the automation, it was the idea that a website could be alive. Not in the "real-time updates" sense, but actually regenerating itself based on what's happening in your life.

How Mine's Different

I took the same core idea but adapted it to what I actually wanted to explore. Erik's running four models in parallel to compare their outputs side-by-side, which is fascinating if you want to see how different AI models interpret the same creative brief.

I went a different route. Instead of comparing models, I wanted to see how different personas could tell the same story. So my setup runs five model variants sequentially:

1. Gemini - Standard generation with creative freedom
2. Poetic - Gemini as a poet who hates unnecessary words
3. Claude Opus - Claude's take with its own chosen persona
4. GPT - OpenAI's interpretation
5. Poetic Hindi - GPT as a Shayar speaking in Hinglish

Every day at 6 AM IST, a GitHub Action kicks off and generates all five versions. The pipeline is straightforward:

1. Collect context - Gather prompts and site structure into generated/context.json
2. Generate - Run each model variant, creating content in generated/[model-name]/
3. Verify - Make sure nothing outside generated/ was touched
4. Archive - Copy all builds to public/builds/YYYY-MM-DD and update the manifest
5. Deploy - Vercel picks it up automatically

The whole thing runs in about 5-7 minutes.

Why This Actually Matters

The real insight here isn't about portfolios or automation. It's about websites that adapt to reality without human interpretation.

Most websites are static snapshots of what someone thought was important when they last had time to update it. The gap between what's actually happening and what the website says grows wider every day until someone manually bridges it.

This pattern flips that. The website becomes a living reflection of actual state, not remembered state.

Think about what this enables:

For SaaS products - Your landing page could regenerate based on what features you actually shipped this week. Not what marketing planned to highlight three months ago. The site reads your changelog, understands what changed, and rewrites the messaging to match. No more "coming soon" badges on features that launched last Tuesday.

For documentation - Docs that rewrite themselves when your API changes. The AI reads your OpenAPI spec, detects breaking changes, and regenerates the affected pages with migration guides. Developers see accurate docs, not outdated examples that reference deprecated endpoints.

For content sites - A news site that doesn't just aggregate headlines but synthesizes them. Every morning, it reads the day's events and regenerates its homepage with context-aware summaries. Not just "what happened" but "why it matters given what happened yesterday."

For e-commerce - Product pages that adapt to inventory, seasonality, and trends. Out of stock? The page doesn't just say "unavailable" - it regenerates to suggest alternatives based on what people who bought this also looked at. Winter coming? The product descriptions shift tone to match the season.

For internal tools - Company dashboards that don't need manual updates. The system reads Slack, GitHub, Linear, and meeting notes, then regenerates the "what's happening" page every morning. Teams see actual progress, not last quarter's roadmap slide.

The pattern is the same everywhere: pull real data, synthesize meaning, regenerate presentation. No human sitting there interpreting what changed and manually updating copy.

Why I'm Writing This

I'm not trying to convince anyone this is the future of web development. Maybe it is, maybe it isn't. But it's a fun experiment that actually solved a real problem for me - keeping my portfolio a little dynamic.

And honestly, there's something satisfying about waking up to a site that updated itself while you were asleep. It feels less like maintaining a website and more like having a small piece of the internet that evolves on its own.

That's kind of the point.