AI agents for marketing can already do most of the work behind a content strategy: researching the industry, extracting narratives from existing content, mapping creators and competitors, and analyzing the target audience. In my experience that's about 80% of the effort. The last 20%, choosing the angle that fits one specific leader at one specific company, still needs human judgment. The best setup today is an expert directing agents, with the agents learning from every decision.
AI agents for marketing can take on most of the work behind a content strategy today. They can research an industry, scrape relevant sources, pull the recurring narratives out of a leader's existing posts, find the creators and adjacent players worth watching, and profile the target audience. That's the bulk of the hours.
What they can't reliably do yet is make the final call: which angle this particular person should own, given their industry, their company and their own career. When I let an agent do the whole thing end to end, the strategy it produces is passable and a little disappointing. When I direct it and make the final calls myself, the results are very different.
This is an active argument inside my own company, so let me lay out both sides.
What AI agents for marketing actually are
An AI assistant answers when you ask. An agent works toward a goal: it can use tools (search, scraping, a spreadsheet, a publishing API), take several steps in a row, check its own output, and remember what it learned last time.
For content marketing, that difference matters. A chat assistant writes a post when you paste in a prompt. An agent can go read your last fifty posts, the posts of your top competitors' executives and the profiles of the people you want to reach, and come back with a view on what's missing.
Here's where I see agents earning their place in content marketing right now:
- Research. Scraping sources, summarizing industry news, collecting what competitors and their leaders are saying.
- Narrative extraction. Reading a leader's existing posts, talks and call notes and pulling out the themes they keep returning to.
- Ecosystem mapping. Finding the creators, analysts and adjacent companies your audience already follows.
- Audience analysis. Looking at what your target buyers actually engage with, as opposed to what they say they want.
- Drafting and repurposing. Turning approved ideas into posts, threads and newsletter sections in the leader's voice.
- Execution. Scheduling and publishing, with a human approving each post.
The first four are strategy work. That's where the interesting question is.
Can an agent set your content strategy? The argument with my CTO
My co-founder, who is our CTO, and I disagree about this.
His position: anything that is text in, text out can eventually be fully automated. If a problem can be written down as text and fed to a model as text, you can get great text back. He has seen this play out at his previous startup, which worked on video production, where more and more of the pipeline could be expressed as text and handed to models.
My position: we're not there yet for content strategy. When I built my own strategy process using an agent, I saw clearly the gap between what the agent produced and what my judgment added. You can build a fully automated basic content strategy today. The result is pretty disappointing.
Where we agree is the part that matters for anyone buying or building with agents today: you need an operator. Someone with real expertise and proprietary insight has to direct the agent and teach it what good output looks like. Whether the agent eventually needs that person less is the open question. For now, I'm driving, and the agents ride along and learn.
Why content strategy is hard to automate: three layers
A good content strategy for a leader sits on three layers at once:
- The industry. What's happening in the market, what the audience worries about, what everyone else is already saying.
- The company. Its stage, its goals (pipeline, hiring, fundraising), its positioning and its competitors.
- The person. Their career arc, what they've achieved, what they've seen that nobody else has, and how they talk.
Two leaders at the same company should have very different strategies, because the third layer is different. A CTO and a CEO at the same startup share layers one and two and almost nothing in layer three.
Agents are good at layers one and two, because those are mostly research. Layer three is where it breaks. The details that make a person worth following are often not written down anywhere, and when they are, an agent can't easily tell which of them matter. I wrote about what happens when you get this wrong in we matched a CEO's voice perfectly: the voice was right and the posts still sounded like every CEO on LinkedIn. And the reason a strategy has to be an intersection of audience and authority is in LinkedIn content strategy.
The 80% AI agents for content strategy can own
Here's how I split the work today. The 80/20 split is my rough sense from doing this repeatedly.
| Strategy task | Agent | Human |
|---|---|---|
| Scrape and summarize industry sources | Owns it | Spot-checks |
| Extract narratives from the leader's existing content | Owns it | Confirms which ones are real |
| Find creators and adjacent players in the space | Owns it | Picks who's relevant |
| Analyze what the target audience engages with | Owns it | Sanity-checks the sample |
| Map competitors' leaders and their angles | Owns it | Reads the shortlist |
| Choose the angle this leader should own | Suggests options | Decides |
| Decide what the leader should never post about | Flags risks | Decides |
| Judge whether a draft sounds like the person | First pass | Final call |
The top half used to take days of a strategist's time. Now an agent does it before the kickoff conversation, which means that conversation starts at a much higher level. Instead of asking a founder "who's your audience?", you can ask something like "your buyers engage far more with customer stories than with product news; does that match what you hear on sales calls?"
The bottom half is the 20%. It's small in hours and it decides everything.
Personalized AI experts: training an agent on your judgment
The most interesting development, to me, is a new kind of product: agents trained on a specific named person's expertise.
My co-founder and I have built these for ourselves. We have agents named after each of us, and we use them daily. Mine is trained on content strategy and company strategy, plus content creation and writing. His is trained on architecture, coding and a set of engineering skills. We use them to apply something close to our own judgment and taste to cases we haven't seen yet.
It's a different model from the software we grew up with. For the last decade or more, SaaS packaged a team's expertise into features. You got exactly what the features allowed. If your case went a little left or right of what the product was designed for, you were stuck. LLMs changed that because they aren't deterministic: they can adapt to a situation nobody designed a feature for.
The way you train one of these experts is with cases. For my strategy agent, that means detailed questions and answers across very different situations: a B2B company versus a B2C one, one industry versus another, and most importantly a specific leader, a CEO or a CRO, with a career that belongs only to them. With enough cases like that, the agent starts to generalize the judgment to cases it hasn't seen.
It will never be exactly the advice I'd give sitting one-on-one with that person. It gets close, and it beats generic advice by a wide margin. It's also a lot cheaper than a consultant billing hundreds of dollars an hour.
This is also where a personalized expert differs from the chatbots built on a famous person's published words. Those can answer questions in someone's style. What they usually lack is tools, and the depth to do a specialized job end to end. A content strategy agent has to research, decide, draft and ship.
How to start using AI agents for content strategy
You don't need to build a named expert to get most of the value. A practical starting point for a founder or a small marketing team:
- Give the agent your raw material. Past posts, call notes, voice memos, your positioning doc. The more of layer three it can see, the less generic its output.
- Run the research as a batch. Ask it to map your industry's narratives, your competitors' leaders and your audience's engagement. Read the output critically.
- Make the angle decision yourself. Pick the two or three themes you want to own. Write down why.
- Write your rules down. What you never post about, words you don't use, claims you can't back up. Agents follow written rules far better than implied ones.
- Let it draft, and correct it in writing. Every edit and every rejected idea is a training case. Feed the reasons back.
- Keep a person on approval. Nothing goes out without the leader seeing it.
If you work in Claude, this can all happen in one place: the agent reads your material, drafts in your voice and, through a LinkedIn MCP server, schedules the posts you approve. Want ideas to test the setup on? Run your positioning through the content ideas tool and see which suggestions an agent would have to fix.
Where this is heading
I don't know yet whether my CTO is right and a fully agentic strategist is a year or two away. What I can see is the direction. Our agents get better every week, because every time I make a call they didn't, that decision becomes part of what they learn from.
That loop is the real advantage, and it only works if an expert is in it. My rule applies here too: you can't lead what you've never touched. Marketing leaders who get their hands dirty with agents now, directing them and correcting them, will be the ones who know where the 80/20 line actually sits in their own business.
