Breyta

Our Ahrefs audit showed 599 errors and a health score of 38. Nine days later: 2 errors and a score of 99. And I did not fix a single one myself. An AI agent we built at Breyta did the work. What it did over those 9 days: → Audited the whole site and found the 599 errors → Opened GitHub PRs with the real fixes: canonicals, meta tags, redirects, broken links, sitemaps, structured data → Cut errors from 599 to 2 → Took the Ahrefs health score from 38 to 99 → Now it checks the site every week and opens new PRs when something breaks I read the PRs and merged the ones I wanted. That was my whole job. A few things people ask me: → Isn’t this too hands-on? No. It writes the fixes as pull requests. You read the diff and merge, so nothing changes on the site until you do. → Is it safe to point at a live site? Yes, because it proposes and you approve. Every change is a PR you see first. → Won’t AI search still skip my pages? It does the AEO cleanup too, so AI search can actually read them. The numbers: 38 to 99 on the Ahrefs health score, 599 errors to 2, in 9 days. Both numbers are Ahrefs, not mine. What you get: ✅ An agent that audits your site every week ✅ Real fixes delivered as GitHub PRs you approve ✅ A free Breyta template to copy and customize, or the hosted app if you just want to use it It’s live now and free to try. Do you want me to send you the link? 1. Connect with me (the DM won’t land otherwise) 2. Comment "SEO FIX"

Every morning, we wake up to one or more small PRs that improve our codebase. They’re opened overnight by an AI coding agent running on a VM we control. The job is simple: find and implement one small worthwhile improvement, run the checks, and only open a PR if the change is small and solid. There is always a feature to ship, a bug to fix, or something more urgent to do. I always seem to postpone these cleanups and improvements. This setup gives that work a place to happen. Most mornings the changes are exactly the kind of work we rarely make time for ourselves: repeated code extracted to a common helper, a small bug in hiding fixed, a simplification or more idiomatic rewrite of a piece of code. One recent PR cleaned up a Firestore subscription path so listener registrations were tracked and removed properly. You can piece something like this together with cron jobs and scripts, but those setups can get fragile fast making it hard to know what actually ran, or whether to trust it. This is exactly the kind of job Breyta is for. Breyta is the reliable orchestration layer around the AI agent: start the job at the right time, run it on the right machine, wait for it to finish, get the result back, and keep the whole thing visible, inspectable and repeatable. If you’re curious how our setup works, I wrote it up in more details here: https://lnkd.in/eSMue7Y7