Nova

Most of what an agency does, day to day, is not strategy. It is a loop:

1. Checking SEO rankings.
2. Monitoring local SEO.
3. Tracking visibility inside AI search.
4. Checking citations and directory listings.
5. Researching what competitors are doing.
6. Coming up with content ideas.
7. Turning those ideas into a content strategy.
8. Running QA across the website.
9. Making small changes to pages.

Every one of those is necessary. Almost none of them need me. They need someone competent doing them consistently — and consistency is exactly where this kind of work fails. Not because it is difficult, but because it never ends, and there is always something more urgent.

So the work slips. Rankings get checked when something feels wrong rather than on a schedule. Citations get audited once, at the start. The content calendar survives about two months. None of that is negligence; it is what happens when the person capable of doing the strategic work is also the person doing the repetitive work.

I built Nova to take the second job away from me.

In practice it now handles roughly 80% of the repetitive work normally carried by a junior strategist, developer or designer. That figure is my estimate from running it, not a measured benchmark. What it buys is not speed — it is that the repetitive work stops competing with the work that actually needs judgement.

Nova is not one large AI answering prompts. It is a set of specialists with narrow jobs, and a project manager that decides who works on what.

• Project Manager — decides what needs doing, in what order, and hands it to the right agent.
• Researcher — gathers the source material: competitors, market activity, what is being asked.
• Content Strategist — turns research into a plan, and the plan into briefs.
• Writer — drafts against the brief.
• Editor — checks the draft against the brief rather than against taste.
• SEO Agent — rankings, local visibility, citations, internal and external links.
• QA Agent — crawls the live site looking for what has broken since the last pass.
• Web Developer — makes the fix, rather than filing a report about it.
• Designer — handles look, layout and imagery.

Most multi-agent systems fail in the same place: step four has forgotten why step two decided what it decided. The specialists are the easy part. Passing context down the chain without losing it is the work.

This is the part that separates Nova from generic AI automation.

Most AI tools start every task from nothing. You paste in context, you get an answer, the context is thrown away, and tomorrow you paste it again. The tool never gets better at working for you specifically.

Nova has a memory. For a law firm, it gradually builds up a picture of the business:

That memory is what makes the difference between an agent that writes a competent article about personal injury and one that writes the article this firm would have written — naming the right attorney, in the right voice, for the right city, without repeating what was published in March.

The longer Nova runs for a business, the more context its agents have, and the less direction they need.

Nova is not a prototype or a demo. It is what I use to operate my own projects and my client websites, in production, today — eight sites at the time of writing, all running on the same platform, which is what makes a fix made once a fix everywhere.

• CarlosArias.com — the site you are reading. Written, checked, repaired and rebuilt by Nova.
• Medellin.co — a city guide of several hundred pages with a self-verifying business directory. This was Nova v2, and where most of the hard lessons came from.
• Private law-firm client — powered by Nova — named on request.