← All posts

Sela Is a Managed, Embedded SEO Agent. Here's What That Means.

Most SEO work arrives as a report. Someone audits your site, sends you a PDF with forty findings, and leaves. Six weeks later half the findings are still open, nobody remembers why the other half mattered, and the next audit starts from zero.

Sela, our SEO and AI-visibility agent, is built the other way around. She is embedded: she lives inside your own AI, where your team already works. And she is managed: we are her manager, and we never stop managing. The audit is the easy part. The value is two conversations that never stop, ours with Sela and yours with Sela.

Embedded: she works where you work

Sela doesn't live on a dashboard you have to remember to log into. She is connected to your team's AI, so anyone can ask her, in plain English, "what should we fix next, and why?" She answers from your own audit and your own task list, not from a generic SEO checklist.

Each fix she recommends arrives as a task with a reason attached, written as a benefit to your business rather than a technical finding. When your team finishes one, they tell her. Nobody takes that on faith: a task closes only after we see the change on the live site.

She advises; your team decides and makes the change. That is deliberate. Your site is yours.

Managed: we are her manager

An AI agent is smart and has no memory of yesterday. Every day is her first day. So the job of managing her is to make sure what she learned yesterday is written down where today's Sela will read it. We do that two ways.

Tool dossiers

Sela works through a stack of tools: search data providers, the AI engines themselves, Search Console, site crawlers. Every one of those tools has quirks, and the quirks change. So every tool has a dossier: one file for how the tool behaves for anyone, and one for our own accounts, settings and gotchas. We keep about 50 of them, and every fact in them carries the date it was last verified.

One example. Early on, when an AI engine answered a question without searching the web, we scored it as a zero: "didn't name the client." That looked like a visibility problem. It wasn't. It was a measurement problem. The rule is now a dated line in that tool's dossier: an answer with no search is excluded, never counted as a zero. Today's Sela reads it before she measures.

Procedures that improve themselves

Everything Sela does more than once is a written procedure, with a check after each step and a separate grader that reviews the run. Across the company we run 187 of them. The SEO audit is one: 46 steps, ending in a short, ranked list of fixes.

The procedures are not fixed. Each run teaches the next one. Sela's SEO audit went through three versions in four days this month: version 1.0, then 1.1 after she audited our own site, then 1.2 after her first client run. On that client run the grader flagged 17 items, and Sela fixed them the same day. A second, fresh agent then checked the client documents and found 14 more errors before the client saw any of them. Three days later the audit grew into a two-week cycle: a quick recheck, alternating with a full audit once a month.

Lessons climb the same way. A mistake becomes a one-line lesson. If it shows up twice, it becomes a standing rule. Every Friday, we review the week's lessons and decide which ones change a procedure.

Two loops, running all the time

Put together, there are two conversations, and neither one ends.

Us and Sela. We are her manager. We set her priorities, review her work, approve anything that touches a client, and turn each mistake into a better procedure or a dated dossier line. She gets better on the job because we write down what the job taught her.

You and Sela. She is your agent. You ask her what to do next and why. She explains, you decide, your team acts, and she checks the result. Your questions and your calls feed back into the task list, so the plan moves with your business instead of with last month's audit.

The two loops meet in the middle. What you ask her this week shows us where her answers fall short. What we fix in her procedures shows up in her next answer to you.

What this doesn't mean, yet

This is early. Sela's SEO work is live with a small number of clients, and I'm not going to make that sound bigger than it is. She doesn't edit your site, and she doesn't run on autopilot: every audit, task and report she gives you is approved by a person here first, and her answers come from that approved material.

Current view, subject to change

I think the agent is not the product. The management is. Plenty of tools can run an SEO audit. Few will still be paying attention, and getting better at your site, in month six. If clients tell me the loop feels like noise rather than help, that's the fact that would change my mind.

Regards,

Charles Stack
Founder, Coworkers.Global

Coworkers.Global is an AI staffing agency. We place managed agents into organizations that need dedicated expert knowledge work. A managed agent is an AI specialist provisioned for a specific role, trained on your context, supervised by a person, and accountable for its output. The first, Alex, evaluates startup business plans for fundability, informed by human expertise and research, and calibrated against real investor decisions. We are early-stage with paying clients, and we still lead with the quality of our judgment rather than customer logos. Your managed AI coworker.
Get new posts by email

What we're learning building a startup with managed agents, plus notes on raising. Weekly newsletter, no spam.