We sell AI automations, so we should run our own company on them. As of this week, a system of 10 agents we call Finessely Engine handles the research and first drafts for all of our sales and marketing.
Why we built it
Growing a services company means being present where buyers talk, publishing useful content every week and following up with every lead. For a small team, that work competes directly with client work. We wanted a system that takes the repetitive parts and leaves the judgment to us.
The ten agents
Four agents work on marketing:
- Site Auditor checks every page of our site for SEO and quality issues and writes a ranked fix plan.
- SEO Optimizer proposes better titles, descriptions and alt text.
- Post Writer drafts LinkedIn posts and X posts and threads.
- Content Repurposer turns one article into posts, a newsletter and video scripts.
Six work on sales:
- Listener finds posts on X and Reddit where people ask for help we can give.
- Intent Scorer rates each post for buying intent and explains why.
- Reply Drafter writes a helpful reply for the high intent ones.
- Lead Prospector builds lead lists that match our ideal customer.
- Outreach Writer drafts a personal first email for each new lead.
- Follow-up Agent sends timed follow ups and stops the moment someone replies.
A person approves everything
No agent publishes or sends anything on its own. Every post, reply and email lands in one approval queue, where we edit, approve or reject it. Only approved drafts go out, and only within low daily limits.
Agents do the research and the first draft. People decide what goes out under our name.
This is slower than full automation, and that is the point. Platforms like X, Reddit and LinkedIn ban spammy automated engagement, and people can tell when a reply was written to sell rather than to help. The review step keeps us useful and keeps our accounts in good standing.
Two models, each where it fits
Claude writes and analyzes most things: audits, SEO, posts, intent scoring and emails. Grok drafts the short social replies. Picking a model per job, rather than one for everything, is a habit we bring to every client project.
What comes next
We will write up what works and what does not as the system runs. If you want the same kind of system for your company, our growth automation service builds it in your accounts, with the same approval rules.