Most people treat AI assistants as fancy autocomplete. Ask a question, get an answer, copy-paste, move on. That's not how this site was built.

Aether Intel is a GitHub Pages publication covering AI agents, business, tools, and safety. Fifty-two long-form articles. A live homepage with a hero slideshow. An article archive. A scrolling news ticker. AI-generated editorial hero images for every piece. Google Analytics, AdSense, Cloudflare, custom domain, full deployment pipeline.

All of it, built and maintained through conversation with Hyperagent — the autonomous AI agent platform from the makers of Airtable.

Here is the workflow, and the seven features that made it possible.

The Workflow

Every article on this site starts the same way. A YouTube transcript from an AI creator or analyst gets uploaded to Hyperagent. From there, a single session handles the entire pipeline:

  • 1
    Mine the transcript Hyperagent reads the raw transcript, identifies the strongest angle, and frames it as an editorial story — not a summary, a structured argument with a hook, a thesis, and evidence.
  • 2
    Write the article Full article HTML: headline, deck, pull quotes, stat grid, data table, related links, SEO tags, canonical URL, Open Graph metadata. Formatted to the site's exact standalone-nav specification.
  • 3
    Generate the hero image A custom 1280×720 editorial image matched to the article's emotional tone — not stock photography, not generic AI art. The AI delusion article got a bedroom at 3am lit by a chatbot screen. The CEO sycophancy piece got a lone executive surrounded by floating "Yes" screens.
  • 4
    Update the homepage The 8-slide hero carousel rotates in the new article. The 24-card article grid gets the new card prepended, with the oldest card dropped. The JavaScript data array stays in sync.
  • 5
    Update the archive and ticker The articles page gets the new card. The scrolling ticker JSON gets the new item. Both updated in the same session.
  • 6
    Commit everything atomically All seven files — article HTML, hero image, updated homepage, articles page, ticker JSON — committed to GitHub in a single operation. One SHA, one message, no partial states.

"A tool waits for you to remember it. An assistant reduces the number of things you have to remember."

— The difference Hyperagent makes in practice

The 7 Features That Made It Possible

01 Persistent Sandbox

Every other AI tool resets between sessions. You rebuild context from scratch every time. Hyperagent has a persistent file system — a real working directory that survives across conversations. Every script built to manage this site, every article draft, every image, lives there across sessions. Pick up a conversation and you pick up the project.

02 GitHub Integration

Not just text files. Hero images as binary WebP, updated homepage HTML, ticker JSON, article HTML — all committed atomically in a single operation. Hyperagent handles the encoding, the API calls, the error handling. No terminal required. No git commands. Just a description of what needs to happen.

03 Image Generation

Every article on this site has a custom 1280×720 editorial hero image generated in the same session where the article was written. Not sourced from stock libraries. Generated, resized to spec, converted to WebP, and committed — automatically, every time.

04 Skills and Memory

The site has a specific article format: standalone nav, hero image first, stat grid, pull quotes, data table, related articles. Hyperagent learned this format and stored it as a skill. Every new article follows it exactly, without needing to describe it again. The same applies to the commit workflow — which files to update, what encoding to use, how to bump the deploy tag. Preferences persist.

05 Multi-Step Reasoning

Publishing a new article is seven tasks that need to happen in the right order, with the right data flowing between them. Hyperagent executes the full pipeline from a single description. Not one step at a time, prompting each one manually — the whole thing, one session.

06 Self-Debugging

When things broke — and they did — Hyperagent diagnosed the root cause, fixed it, and put a prevention rule in place. When 7 orphaned HTML divs were corrupting the homepage slideshow, it found them, removed them, and established a rule: homepage HTML is never modified by automated scripts. When files were committed with wrong encoding and the site went white-screen, it identified the mismatch, re-committed all seven files correctly, and wrote a verification script to catch the same error in the future.

"The tool gets more reliable the longer you use it. Every mistake becomes a rule. Every rule prevents the next mistake."

— After fixing the encoding bug that broke the site

07 Connected Integrations

Gmail. GitHub. YouTube transcript processing. Image generation APIs. Hyperagent is not a standalone chatbot — it is a connected agent that moves data between systems on your behalf. One user used it to autonomously book a performance venue for his daughter — Hyperagent found open mic nights, created the contact list, drafted the emails, connected to Gmail, sent them, and confirmed the booking. Same architecture, different use case.

Before and After

Task Before Hyperagent With Hyperagent
Publish a new article Write → hand to dev → wait → review → revisions → deploy. Days. Upload transcript → describe angle → done. One session.
Hero image Brief a designer, wait, review, export, upload. £80–£200. Generated, resized, committed. Same session as the article.
Homepage update Developer ticket. Could be days. Part of the article publish pipeline. Automatic.
Debugging a broken site Stack Overflow, developer, waiting, ticket queue. Root cause identified in minutes. Fix committed. Prevention rule written.
Cross-article consistency Manual check. Easily missed. Flagged automatically. Fixed in one commit.

What This Replaced

For a site at this scale — 52 articles, custom images, GitHub deployment, live homepage management — the standard team would be:

  • Developer — GitHub commits, HTML structure, deploy pipeline
  • Designer — hero images, visual consistency
  • Writer / editor — article structure, formatting, SEO
  • DevOps / project manager — keeping 52 articles, 24 homepage cards, and 8 slideshow slots in sync

All of that collapsed into one Hyperagent subscription and one person describing what they want.

"I built a team of AI agents that brings any startup idea to life... Rowan goes deep on the problem space, delivers non-obvious insights, and spins up sub-agents."

— Hyperagent demo, official channel. The same architecture that runs this site.

The Honest Part

It was not always smooth. There were broken commits, a white-screen site crash from an encoding bug, a homepage slideshow corrupted by orphaned HTML. But every time something broke, the diagnostic process was a conversation — not a GitHub issue or a developer call. And every fix came with a prevention rule built into the next session's workflow.

That compounding reliability is the part that doesn't show up in demos. After enough sessions, the agent knows the site better than most human developers would, because it has seen every failure mode and adapted to it.

Try It

If you are a solo creator, solo founder, or anyone running a content operation without a team — Hyperagent is what closes the gap between having an idea and having a real thing in the world.

You are reading the result. Every article on this site. Every image. Every line of HTML. Built in Hyperagent chat.

→ hyperagent.com