Create. Capture. Connect.The three loops an AI agent runs on LinkedIn.

Apex, running LinkedIn end to end

Most people connect an AI agent to LinkedIn and get a pile of tools.

Search. Enrich. Comment. Message. Thirty-six callable functions and no opinion about what to do with them.

Tools are not the product. Recipes are.

Apex is two things: a governed LinkedIn MCP, and the proven GTM recipes that run on it. The recipes compress into three loops: create demand with posts grounded in what's working now, capture demand into approved pipeline every day, and connect with the next buyers where their attention already is.

Every loop is autonomous about finding and drafting. No loop is autonomous about sending. That difference is the whole product.

Why loops, not tools

A tool answers “what can the agent call?” A loop answers “what happens every day?”

Most outbound systems start with a list — a CSV, an export, a scraped search that somebody still has to clean, enrich, personalize, approve, and remember not to contact again next week. Most AI writing tools start with a blank prompt. Most outreach starts too cold.

The loops start somewhere better: with a signal. What's working in your niche. Who's showing intent today. Where a prospect's attention already is. The agent handles the messy middle; you handle the final judgment.

Loop 01 — Create

The Lead Magnet Post Generator

LinkedIn content doesn't exist in a vacuum. Every day, people are testing hooks, offers, formats, and lead magnets in public. Some posts get ignored; some collect hundreds of comments. The loop starts by studying the lead-magnet posts working in your market right now — not random viral posts, relevant ones.

Then it studies you. Your last 20 posts become a style brief: sentence length, tone, formatting, how salesy you sound. “Write in my style” is not enough — the draft has to survive the only test that matters: would I actually publish this?

Then it drafts the post — and the campaign around the post. The first comment, the auto-DM copy, the comment keyword, the automation settings. A lead magnet post is not just content; it's a demand-capture system, and the agent prepares the whole thing.

Loop 02 — Capture

Daily Saved Search Outreach

LinkedIn is not static. People change jobs, announce pain, comment on competitor posts, show buying intent in public — and then the signal disappears into the feed. A static list misses the point. The better primitive is a saved search that runs every day, described in plain English: “founders posting about hiring SDRs”, “agency owners whose posts are getting 30+ comments.”

Search creates candidates. Filtering creates judgment — out go competitors, existing customers, anyone contacted recently, anyone on the do-not-engage list. Enrichment turns a name into a reason: who they are, what they posted, why today, what the opener should be.

Then the loop stops and asks you. The shortlist arrives with the fit, the source signal, and the proposed message. Approval isn't friction — it's the control point that keeps quality high. Most automation products get lazy exactly here. Only approved actions run, and every send is logged: who, why, which search produced them, who approved it.

Loop 03 — Connect

Rooms They're Already In

Cold outreach usually starts too cold. A profile tells you who someone says they are. Their recent activity tells you what they're actually paying attention to — which posts they comment on, what they react to, which conversations keep pulling them back. Those are the rooms they're already in.

Bad outreach says: “Saw your profile, thought I'd reach out.” Better outreach says: “Saw your comment on the post about SDR hiring — liked the point about reply quality dropping when teams over-automate. Are you solving that with process or tooling?” That opener only exists because the agent found the room.

Most outbound tries to create context from nothing. This loop borrows context that already exists — you're not interrupting cold, you're entering a conversation where the person already showed up.

The gate

All three loops share one governance layer, and it's not optional. One approval queue. Shared daily budgets — every write the agent makes draws from the same limits as your automations. A do-not-engage list that everything respects. Pacing calibrated from millions of real LinkedIn actions, because burst behavior is what gets accounts flagged. And an audit trail that answers: what ran, what's staged, why something was refused, who approved what.

The loops are autonomous about finding and drafting. Never about sending.

The principle

Do not give agents random tools. Give them recipes.

A generic LinkedIn API exposes actions. Apex gives the agent the workflow around the actions — signals, enrichment, suppression, approval, budgets, logging. That's the difference between plumbing and an operating system.

Create demand. Capture it. Connect with the next buyers.

Three loops. One operator. Every send approved.