LinkedIn automation tools never made sense to me. Usually you don't pick the best one, you pick the one that sucks the least. Too stiff, too expensive, badly integrated, or all three at once.
And they all push you toward the same instinct: send more. More messages, more follow-ups, more cadence until someone replies. On LinkedIn that's how you lose people. The best outreach I've done here mostly came down to not sending.
Why sequences don't belong on LinkedIn
Sequences work fine for email. You send messages over a period of time, some standalone, some in a thread, so that if the person missed the first one they hopefully catch the next. You keep going until they reply, block, or opt out. It all rests on one quiet assumption: the person can't see the other ten near-identical emails you fired off to get their attention.
On LinkedIn they see all of it. The messages, the cadence, the pressure, sitting in one window. The moment it's visible, the premise falls apart. It's one thing to open a chat and see five messages spread across six months. It's another thing to see five from the past week.
So the conclusion I keep coming back to: LinkedIn should be fluid, and sequences remove exactly that. Every LinkedIn tool is sequences-first, because LinkedIn outreach grew straight out of email outreach tools. You inherited the strategy whether you needed it or not.
It comes down to timing
When people argue against sequences they reach for "build relationships, not campaigns." That's true, but it's soft, and it doesn't explain what's missing.
A sequence can't be two things at once: active and reactive.
Reactive is the easy half. The person posts something and you respond. A signal points at you, you answer.
Active is the harder, more valuable half. Nobody pinged you. You go and find a signal in their world and move on it first: their company raised, or shipped, or there's some news or event. The last time I did this I saw a post from someone on a team talking about a topic I work in, and I used it to open a conversation with the leadership at that company. The signal came from a junior person; I spent it on the decision-maker.
A campaign is static. It can't do either, because it was written before any of this happened.
Why static is the problem
You might push back: sequences aren't that dumb, they have data. They know if someone opened, clicked, replied. They can branch.
Sure. But you define every branch before you've even met the person. You're never going to anticipate every scenario in advance, because the real world doesn't arrive in pre-enumerated branches. The webinar they're about to announce, the role they're about to change, the thing their CEO is about to post: none of it exists yet when you're building the flow.
The agent decides at run time instead of design time, with the real context in front of it.
There's a fair boundary here. If you're using LinkedIn as an advertising channel, blasting the same message at everyone and hoping a few get hooked, then static is fine. You're not pretending it's personal. But that's a marketing function. Sales is mostly relationship-building and trust, and spam is a bad way to start either one. A static sequence on a sales channel is a trust violation dressed up as a relationship.
And LinkedIn is uniquely bad for it, for the reason from earlier: the medium exposes the machine. The person can watch you not paying attention to them.
What you need
So in the AI era, the tools that win are the ones that bend. If they won't, we'll have agents do the work instead, and save the subscription money while we're at it.
Three pieces:
- A system of record — usually a CRM.
- A Hermes agent.
- The LinkedIn API.
That's the plumbing, and it's the least interesting part. What's interesting is what stops happening.
What it looks like
No queue of scheduled messages. No "day 3, day 7, day 14." The agent just watches and waits for a reason.
Someone comments on your post. It sees the comment, checks the CRM, notices they're at an account you care about, and drafts a connection note that references what they said. Not "Hi {first_name}, I see we're both in SaaS."
A conversation goes quiet. A sequence tool would fire message 3 of 5 right on schedule. The agent does nothing, until that person changes jobs or posts something worth reacting to. Then it brings them back to you with a reason to reach out.
You ask it who's worth a touch this week, and instead of reading a drip calendar it reads the relationships, then hands you the three people where the timing changed.
It stops feeling like a tool when you can just talk to it. "Share my last post with these twelve people and ask for a repost," and it's done, with the CRM updating itself while it's at it. No separate step, no logging it after the fact. The one-off stuff you'd normally do by hand at 11pm becomes a sentence you type once.
And if you really want to go there, you can get on a live call with it. Talk to it like a coworker: "who reached out this week, anyone worth a call, draft me three follow-ups before I log off," and it handles it. At that point it doesn't feel like automation. It feels like a coworker who happens to live in your CRM.
"Isn't this just smarter spam?"
Your skeptical reader is already thinking it, so let's put it on the table: before, I got an obvious template and could ignore it. Now an agent reads my post and sends me something that looks like a human paid attention, but nobody did. A model did it at scale. Isn't that worse?
It's a good objection. The difference between an agent and a campaign is that the agent can and will stop.
If there's nothing to react to and nothing to be proactive about, it doesn't message. A campaign keeps firing regardless; that's what it is. Restraint is the tell. Spam sends whether or not there's a reason to, and an agent willing to send nothing is doing the opposite.
"But there's always some pretext," you say. "Somebody always posted something." True, which is exactly why you get to define what counts. You decide which signals are worth acting on and which ones lie. That bar lives in the system because you put it there. A traditional tool can only offer you the dropdowns it shipped with; an agent is customizable down to that level.
One I use: don't treat someone's marketing posts as a trigger. Say a person announces a webinar on "the mistakes companies make adopting AI in the enterprise," and you happen to sell AI tooling. Hand that post to a naive personalization engine and it reads the topic as the person's pain, so you end up messaging the expert who's about to teach the webinar as if they're the one struggling with the problem. The message is built on an inverted assumption, and the prospect can see exactly which dumb signal you pulled it from.
A dropdown menu in some SaaS can't express "ignore the posts where they're broadcasting." An agent can. The work moves up a level, from writing the copy to deciding which signals the system should trust.
The point
Every LinkedIn tool I've used forces its process onto me, because it inherited that process from email and never questioned it. An agent does the reverse. It bends to how you sell: active when there's an opening, reactive when they reach out, quiet when there's nothing to say, tuned to the signals you trust instead of the ones a template treats as equal.
That's the difference. A faster sequence is still a sequence. What you build instead knows when to send nothing, and that's why it reads like an assistant instead of a script.