I have a LinkedIn automation tool in my stack right now. It does roughly what it promises, and I still find it rigid. I write the message, I load the list, I set the cadence, and the personalization reaches about as far as a first name and a company name. For years I never questioned the one part underneath all of that: where the list came from. The list came from somewhere else, on its own invoice.
That second invoice is what I want to talk about. It is quietly one of the larger line items in an outreach stack; nobody argues about it because it looks like infrastructure, and it has just become optional.
The morning you keep not having
If you run social for a business, you have had this argument with yourself. Outreach works. You have watched one conversation in the DMs do more for pipeline than a strong month of posting. And then Monday lands with a content calendar, three approvals waiting, a client asking why reach is down, and a comment thread somebody has to answer before lunch. Outreach slides, because doing it properly for ten people costs a morning you do not have.
So you look at buying your way out of the problem, and the pricing page tells you what the category thinks you are. Outreach software is built for sales teams with sales budgets, and it assumes you arrive holding a list of people. You do not have a list. Getting one means a contact database on a second subscription, which means you are now paying two vendors to send messages you could write yourself if you had the hours.
I have watched social pros reach that point and quietly decide outreach is a sales function. It is not. It is the part of social that a client can actually see working.
What changed underneath the stack
Taylor Logue runs Taylored Marketing, doing social and demand generation for founder-led businesses. Her first version of this looked like what you would build: Claude writing the messages, n8n connecting the steps, and Apollo underneath supplying the contacts. It worked. She retired it anyway, a few weeks ago, and the reason is the interesting part.
She moved the whole thing into a computer-use agent, the kind that drives your screen and hands control back when it needs you. Hers costs thirty dollars a month. And it finds the people itself. “With this, you don’t even have to have an Apollo, ZoomInfo, or whatever database you wanna use,” she says. The database did not get cheaper. It stopped being a separate thing you buy.
The test she trusts is a client campaign reaching out to affiliates and content creators for a product brand. Twenty connection requests a day, every weekday, for about a month. It has not run out of matching people. She says she keeps waiting for the morning it tells her the pool is empty, and that morning has not arrived.
The mechanics, so you can build it this week
The automation is scheduled, not triggered. Taylor has her runs stacked early in the morning, back to back, so the work is finished before she opens her laptop. Her instruction to the agent was close to plain English: create an automation that produces the day’s messages every morning, and put them in a CSV so she can do the sending manually.
What arrives is a spreadsheet, and the column set is the whole design:
- Name, title and company, so you can sanity-check the match in two seconds
- A niche signal, meaning the agent’s stated reason for surfacing this person
- A suggested connection note, written against her brand voice
- The profile URL, so sending is a click and a paste
The messaging quality does not come from the agent. It comes from two briefs Taylor wrote once and keeps in Claude as Skills: one describing her voice and what she represents, one describing her ideal client and the services she offers. For client campaigns the same material lives in a sheet with five named personas. She is explicit about why she keeps the stack this small. Teaching a new platform about your business is, in her words, the most time-consuming thing, so she would rather run two or three tools well than twenty badly.
Then there is the step that looks like she did not finish automating. The agent will send the connection requests itself. She will not let it. “I’d rather not risk my LinkedIn account or my client’s to have it done automatically.” So the workflow ends at a CSV, and she clicks connect by hand, which costs her a few minutes and buys her a review of every message before it reaches a person.
Where it breaks, and what it costs you
Two honest limitations, because this is new enough that nobody should sell it to you clean.
The first is the brief. This agent does not ask clarifying questions the way Claude does. Give it a thin prompt, and it will work for ten minutes, hand you a list built on nothing, and then take another twenty while it redoes the job properly. The models we have been using for two years trained us to start a conversation and refine as we go. This one takes what you gave it and runs. Front-load everything, or pay the half hour.
The second is provenance. Taylor is straightforward that she does not know how the agent is finding contact information, only that it keeps working. If you are doing this for a client in a regulated sector, or anywhere GDPR applies, that is a question to answer before you scale it rather than after. I would rather name that now than have you discover it in a procurement review.
Her own read on fit is narrow, and I think it is right. This suits solo operators and small teams. She does not expect enterprises to move, because they are already committed to Claude or Copilot, and she expects those assistants to ship the same capability before long.
The part worth keeping when the tool changes
Taylor has rebuilt this workflow twice in a year, and both times the tools were replaced while the thinking survived. Her voice brief and her ideal-client brief moved from one stack to the next without being rewritten. That is the asset. The agent is rented.
So when you look at the next AI for social media tool that promises to do your outreach, ask a different question than the one the pricing page invites. Not whether it can send the messages, because most of them can. Ask which layer of your current stack it makes unnecessary, and ask what you would still own if you cancelled it on Friday. If the answer is a written definition of your voice and your ideal client that any tool could read, you are building something. If the answer is a seat and a saved list inside somebody’s platform, you are renting the whole thing.
Taylor walks through the daily run, the CSV and the check she makes before sending in episode 13 of The AI Social Playbook, with her screen up the whole way. Start with the two briefs. Write them this week, before you pick the tool.
