I came up in marketing automation systems. Scoring models, nurture tracks, routing rules, the big canvas of boxes and arrows. I was obsessed with Marketo and Pardot, and with pushing those tools as far as they'd go.

What I got good at was tying those systems to how a business grows. In those early programs I designed the workflow and the software ran it, and any decision inside it came from a person on the team, usually me.

Shout out to all the OGs that remember this!

About two years ago at Webflow, I built an AEO audit tool in one night. It scraped thousands of sources plus your site, scored the site and handed back on-site fixes. That was a basic agent that did one job, and it felt like a big deal.

These days marketers can build their own tools, the way I did that night, and the tools are far more capable than they were back then. Some systems go further and close the loop: they read a signal, propose a workflow and do the work once you approve it. Then they report back on what happened, and that result feeds the next round.

I build the closed-loop kind for clients. Profound has built this directly into their product, and I've worked with their team for about two years. That pairing of systems and growth is worth a lot more now. A marketer who understands both can build what they used to have to ask for, like the customer research they never had time for or the tool engineering never prioritized.

Where the work starts

I spend my days in AI search, so the signal I think about most is what answer engines tell buyers about a company. The same idea works anywhere a signal moves: pricing page conversion slips, off-site AEO citations start to decay, sales hears the same objection call after call, usage dips the way it does before an account churns.

I'd say marketing automation now includes any task where software interprets a signal, proposes a workflow and executes it within limits a person set.

Out of the box, insanely impactful Profound agents

I like that software takes the interpreting and the executing, because that leaves more room for people to decide what's worth doing and make it good.

Tools that help figure out what to work on are half of this. The other half is building the tools yourself, which is the part I like more as a creative disguised as a growth quant guy.

Building your own tools

A few months ago, most of the agent work I saw was one prompt doing one task. Now a marketer can build a working app or a set of agents without waiting on engineering.

A recent app of mine takes a single webinar, as a YouTube link or a transcript file, and returns three follow-up emails, two blog posts that link back to the recording and a LinkedIn post. First it pulls a structured brief out of the transcript, then it writes each asset from that brief only, using claims the transcript supports.

It took me about four hours and six prompts to build, integrations included: one long prompt describing the whole product, a few short corrections, and a last one that tested it end to end and caught three bugs. I use it on my own content and I've started deploying it for clients.

The app runs the workflow I designed, and I still decide what's worth making. I also built the tool myself, without needing a spot on anyone's roadmap.

Then you can hook a tool like that up to the systems where your data lives. That's what MCP, the Model Context Protocol, is for: it lets an AI assistant read from your tools and act in them. You only get what each vendor chooses to open up, and only what you've given it permission to touch. But plug in three or four of them and the tool you built is working off what's happening in the business right now.

Making one webinar evergreen used to pull in product marketing, content, creative and marketing ops. One marketer can build the tool for it in an afternoon, and the time saved goes to the creative part: making the webinar itself as good as I can, and deciding what goes in it and how it gets distributed.

I'd want a person reading the output against the source before it ships, and I'd want every tool to have one owner who keeps it connected and fixes it when a run breaks.

Every discipline has a tax

This is why I'd stop filing automation under lifecycle or ops. Every marketing discipline carries a tax, the work nobody was hired to do that eats the week anyway.

  • Content: finding which pages carry stale claims, and reformatting one piece for five channels.

  • Paid: pulling performance across accounts and building variant after variant.

  • Product marketing: tagging sales calls for objections, and logging what competitors changed on their pricing pages.

  • PR and comms: compiling which publications the answer engines cite for your category.

  • Ops and analytics: stitching the weekly report together from five tools.

I'd automate pulling the paid performance and keep a person deciding that a creative has worn out and what should replace it.

I'd want the freed-up hours going to customers, to an original argument you finally have room to develop, and to the ambitious test that never made it out of the planning doc.

It also makes experience worth more. Knowing your buyers, spotting a weak claim and understanding why a campaign failed are what you build into the tool, so a marketer with ten years of judgment can turn it into something that runs.

The job gets better too. A marketer can go from noticing something to shipping a response without waiting on three other teams. They can design the experiment, build the tool and watch the result come in. That's a better job than the one I started in, by a lot.

How one vendor has set this up

Profound is innovating quickest for cross-discipline marketing agents. They started as the place you went to see how you showed up in ChatGPT and the other answer engines, a very good dashboard with an impressive dataset behind it. They've moved well past that since, and their ship speed has been insane to watch.

First, they let the data out of the dashboard. Their MCP server puts your visibility, sentiment and citation data inside any AI tool that supports MCP. From there you can also start Profound Agents, build and publish new ones, and check on runs, and Profound's own write-up says an agent built once can be shared with the whole team that way.

Then they went after the work itself. AI Marketer, which launched in July as Aim, watches your AI search data and surfaces opportunities, and the ones you accept become projects, per the product page. Each project comes with a brief you can edit. You sign off on the plan, it builds the agent that does the work, and it reports on what moved.

I'd expect the rest of the market to move this way: the data in the tools I already use, software that tells me what it thinks we should do and then does it, and me approving it. Profound has shipped each piece.

The order of those steps matters to me because of the work I'm doing right now. So much of AEO is off-site, which means getting the brand into the sources the answer engines actually cite, and plenty of people only sell on-site work because it's the part you can control.

I'm building an agent that reaches out to the people behind those sources and negotiates the citation. It's the slowest, most manual part of the programs I run for clients, which is why I'm automating it. I give it the targets, the product proof, a budget and hard limits, and I approve the spend and the final terms.

It only works if the research comes first. The agent has to know which sources the answers lean on before it contacts anyone, and that's what citation data like Profound's is for. Point it at the wrong roundup and you've spent a week winning a citation that moves nothing. I'd also have a person read every message before it goes out, because it carries the client's name.

My bar is the same for any vendor, Profound included. I want my data reachable through an open standard, and actions I can call from my own tools. I want a person signing off. And before I expand anything, I want to see a real run: how long it took, what the reviewer changed and what happened after.

What I'd leave to people

Everything below is work I love.

Positioning is the big one. Profound's Context Manager distills your knowledge base files, meeting transcripts and communication threads into brand context, per the Series D announcement, and the AI Marketer works from it. It's on Profound's Enterprise and Agency plans. So whatever your positioning doc says is about to get said a lot.

A good writer patches vague positioning as they go. Software repeats the mush at volume, so the point of view has to be sharp before it goes in. Profound says the AI Marketer adjusts when your positioning changes, so I'd treat that doc as the one place to update. The people who can say what the company believes, and what it would never say, are the ones I'd want writing it.

Proof works the same way. A system can pull the customer data and run the analysis. Somebody still has to sit across from the customer, earn the story and put their name behind the claim.

Taste is where I get nervous and excited at the same time. When making variants gets cheap, more of the job becomes picking, and you learn to pick by making things and watching some of them flop. If a junior marketer spends their first few years clicking approve, I don't know where their judgment is supposed to come from.

The upside is that a junior can now take a swing at ambitious work in their first year with almost nobody to wait on. I'd set that up on purpose: juniors make work from scratch, ship small tests, explain what they threw out and why, and study what happened.

Then there's saying no. These systems will surface more opportunities than any team should chase, and some of those gaps belong to buyers you don't want. I'd leave the call on which ones to walk past with a person.

How I'd judge whether it's working

I'd keep two scorecards and I wouldn't blend them.

The first is production. The number I'd track is total cost per accepted deliverable, at a quality bar you agreed on up front. Count the license, setup, integrations, the hours spent keeping the context current, review time and rework. Put that next to what the same deliverable costs from your team or your agency today, and track turnaround time too. If those improve, you have an operational win. That's worth having, and it tells you nothing yet about growth.

The second is commercial: qualified pipeline, conversion, retention, or whatever business outcome the workflow is supposed to improve. Get a baseline before you turn anything on, pick the segment, fix the window and use a metric that fits the workflow.

In AI search, one page update can affect both your target questions and your comparison group, so I'd treat the comparison as directional unless you can show the two are really separate. And two numbers going up at the same time doesn't mean one caused the other.

What to do now?

I'd run one pilot before I touched a budget line or an org chart. Pick a single recurring workflow and write down what it costs today, how long it takes and how good the output is.

Before you start, decide whether the pilot is testing efficiency, growth or both, and write down what earns expansion. Setting the rule first keeps you from talking yourself into the result. For efficiency, how much cheaper or faster does it have to be, at what quality bar, with how many review hours? For growth, which commercial number has to move, and by how much?

Run it for a fixed window. If it clears the bar, expand it. If it doesn't, change the setup or shut it off.

What this means for 2027 planning

For planning, I'd ask which recurring decisions and tasks a system can reliably take on, and what proof I'd need for each. I'd also ask which recurring work the team should build its own reusable tool for. Those two lists are the automation roadmap, and they reach well past the nurture programs the phrase used to cover.

You can build things now that used to need a bigger team, a bigger budget or a spot on engineering's roadmap, so I'd start with a small, testable version of the idea you've been sitting on the longest.

I think about songwriting a lot when I think about marketing. The storytelling is the part I love. I built my career in automation and systems, but what always got me going was tying that work to the business and finding a creative way to bring it to life. The tools are now good enough to close the loop on the automation, which gives us the time back to spend on the story.

JG

Keep Reading