Always-on agents · In real life

Always-on agents for marketing teams: listening, publishing, reporting

I sell an always-on agent to marketing teams, so read this with that in mind. It is also why I have spent a year watching which marketing jobs an agent genuinely does better at 3am, and which ones I would never take away from a person.

In short

Marketing has three jobs with a clock and a clear definition of done: listening to what an audience responds to, publishing on a cadence, and reporting what each post earned. Those are where an always-on agent beats a person who sleeps. What the brand says, anything that cannot be undone, and anything that spends stay with a human. I run my company on that split.

Part of the Always-on agents series, post 26 of 33.

Every vendor in this market, mine included, will tell you an agent can run your marketing. That is the wrong claim, and I say so as someone whose product is one of these agents. An agent runs the parts of marketing that have a clock and a finish line. It should not run the parts that are a judgement about who you are. The useful work is drawing that line precisely, so here is where I draw it after a year of running Leo for brands and a handful of smaller agents for my own company.

Three jobs with a clock

In the first post of this series I gave three tests for an always-on agent: it holds a goal, it starts itself, it acts and reports. Marketing has three jobs that pass all three without any strain, and they share a shape. Each has a trigger that is not a person, each has an obvious definition of done, and each gets worse when a human does it in batches on Monday morning.

Listening

An audience does not respond to things on your schedule. A competitor posts at 11pm, a trend peaks over a weekend, a customer complains on a Tuesday afternoon. A person checks three times a day at best. An agent checks every hour and never gets bored of it. This is the job Leo was built around: it watches what the audience you have described, your ideal customer profile, is actually responding to, and treats that as the signal for everything downstream. The free, do-it-yourself version of the listening half is in the competitor and brand watch post, and it is a good first agent precisely because it reads and never writes.

The market agrees on this one. Hootsuite introduced Wisdom on 24 June 2026 as a social-first agent that turns live social signals into contextual intelligence and action, with its Talkwalker listening data flowing into every answer, and it is shipping MCP connectors so other agents can read that social context. Listening is the job every vendor has decided an agent should own.

Publishing

Cadence is the lever a marketing team controls most directly and keeps most badly. Posting daily is a different volume of work from posting weekly, and when the work is done by a person it is the first thing to slip in a busy week. An agent does not have busy weeks. Leo's version of this is an hourly loop: find a fresh trend, match a clip from footage the brand already has, edit it, and book it onto the calendar on the next open day. Instagram Reels is live, YouTube Shorts is in beta for invited accounts, and the accounts are connected with a one-click authorisation rather than a shared password.

The rule I am proudest of in that loop is the one that says no. If nothing in the library matches what the audience is responding to, nothing is posted. No match, no post, never filler. A publishing agent without a refusal rule is a spam machine with a nicer dashboard, and I would not run one.

Reporting

Reporting is the job marketing teams most want to hand off and most often do by hand on a Friday. It has the cleanest definition of done of the three: for every post, what did it earn, and which audience did it earn it from. Leo tracks the metrics on every post it publishes, ranks which audience profiles are responding, and drops a click-tracking link into the first comment so a post can be tied to a result rather than a view count. HubSpot's Agent Hub does the equivalent inside the CRM with agents for customer conversations, prospecting research and data quality, each configured once and then left to run.

The honest caveat on reporting: an agent can tell you what happened and rank it. It cannot tell you why a result matters to your business this quarter. That sentence is still yours.

Where I keep a human

Three things never go to an agent on my watch, and the reason is not that models are bad at them. It is that the cost of being wrong is not recoverable.

  1. What the brand says. An agent can draft a caption, pick a clip and propose a hook. The decision that this is what we say, in our voice, about our product, stays with a person. In Leo that is why the footage comes from the brand's own library rather than being generated: the agent chooses from things you already said.
  2. Anything that cannot be undone. Publishing a post is reversible; I can delete it. Replying to a customer in public, deleting a channel, changing a bio, or sending an email to a list is not. My rule from the permissions post applies: if I cannot undo it from my phone in a minute, a human does it.
  3. Anything that spends. Boosting a post, buying placement, paying a creator. An agent can recommend and prepare the brief. It does not hold the card. I keep a hard cap on every model budget my own agents use for the same reason.

The split, on one table

JobWho owns itWhy
Watching the audience and competitors, hourlyAgentTrigger is the world, not a person; reading only
Choosing which moment in the library fits a trendAgent, with a refusal ruleJudgement that changes hourly; recoverable if wrong
Cutting, captioning and booking the postAgentClear definition of done; same quality at 3am
Deciding what the brand stands for and saysHumanNot a clock job; the cost of wrong is the brand
Replying in public, changing account settingsHumanCannot be undone
Spending on boosts, placements or creatorsHuman, with an agent's briefMoney needs a cap and a signature
Weekly report of what each post earnedAgent collects, human readsCollection is mechanical; meaning is not

What it looks like when the split works

I will not quote engagement numbers I cannot source, so here are the two I can. Our product whitepaper reports a video reaching 4k views in its first few hours and a new Instagram page built to a 3k audience, with the editing done by agents in place of five desktop tools. Those are our own claims about our own product and you should read them as such. What I can say from the inside is that the shape of the team changes before the numbers do: the person who used to spend Monday scheduling spends it deciding, and the agent spends the week doing.

That is the same shape I described for my own company in the first post. My publishing job and my site watchdog are not clever. They are reliable, they run at 09:05 and 09:40 every day, and the one morning one of them failed taught me more about agents than any launch keynote did. Marketing agents are the same bargain: reliability you can forget about, bought with rules you have to write down first.

Five questions for any vendor, including me

  1. What does it do when nobody is typing? If the answer is nothing, it is a chatbot with a marketing budget. See agent, chatbot or automation.
  2. What does it refuse to do? Ask for the rule that stops it posting. If there is none, you will find out what filler looks like on your own feed.
  3. Whose identity does it act under? Yours, a shared login, or its own connection you can revoke. Only the last one is acceptable to me.
  4. Can I undo what it did, and how fast? Every action should map to a reversal you can perform without a support ticket.
  5. Where does the content come from? Your own footage and words, or something generated to fill a slot. The answer tells you whether the brand stays yours.

If you want to see how Leo answers those five, the always-on page and the pricing page are written to answer exactly them. If you would rather build the listening half yourself first, the open-source comparison and the brand watch build are where I would start.

Sources

  1. Leo AI, Always-on: a channel that runs itself (accessed October 20, 2026)
  2. Leo AI, Pricing and packaging (accessed October 20, 2026)
  3. Leo AI, Product whitepaper: agentic social media (accessed October 20, 2026)
  4. Hootsuite, Hootsuite rebuilds for the AI era, introducing Wisdom (accessed October 20, 2026)
  5. HubSpot Knowledge Base, Understand Agent Hub (accessed October 20, 2026)

Written by Lera Leonteva. Leo AI sells the publishing agent described here, so I have a commercial interest in this subject.

Frequently asked questions

Which marketing jobs should an always-on agent do first?

Listening. It reads and never writes, so the cost of a mistake is a bad summary rather than a bad post. Once you trust what it sees, publishing on a cadence and collecting the numbers for the report are the next two. All three have a clock and a clear definition of done.

Should an agent write our social posts?

It can draft, select and schedule. In my view the decision about what the brand says stays with a person, and the safest agents choose from things you have already said, your own footage and your own words, rather than generating new claims to fill a slot.

What is the difference between a social media scheduler and an always-on agent?

A scheduler posts what you queued, when you queued it. An always-on agent watches what the audience is responding to, decides what to make from your library and whether to post at all, and books the result itself. The decision is the difference, and so is the refusal rule.

How do always-on agents like Hootsuite Wisdom and HubSpot's agents fit in?

Hootsuite's Wisdom (24 June 2026) owns listening and planning on top of Talkwalker data, and exposes MCP connectors so other agents can read social context. HubSpot's Agent Hub runs customer, prospecting and data agents inside the CRM. Leo sits in the publishing seat: footage in, posts out, with the numbers attached.

What should never be automated in marketing?

Anything you cannot undo and anything that spends. Public replies, account changes, emails to a list, boosts and creator payments all stay with a human in my company, however good the model.

Lera Leonteva
Lera Leonteva
CEO & Founder, Leo AI

Lera Leonteva is the founder and CEO of Leo AI, where she works on turning long-form footage into short-form video that earns attention on its own merits. She writes Video Scaling Strategies, taking one piece of content that worked and one that did not, and using what the market is doing to explain the gap.

The always-on agent I run for social video

Leo watches what your audience is responding to, picks a clip from footage you already have, edits it and books it on the calendar. Every hour, without being asked. It is the agent this whole series grew out of.

Keep going with the series

Every post in the Always-on agents series is on one page, newest first. If you would rather see the agent I run for social video, here it is.