Marketing agents are coming, and paid media looks like the first real category of B2B marketing agents. Most of marketing is taste-based. Paid is closer to the domains where agents already work. Vendors are moving on both sides of the AI x GTM Tech Triangle: buy the agent, or build your own and buy the Legos. Today the agents recommend and humans approve. The most ambitious vendors and builders will push to full autonomy.

Last month I wrote that marketing agents are coming.

We’ve seen agents take off in coding, support, legal, and many verticals.

The GTM agent wave is inevitable, but hasn’t hit land yet. Within GTM, sales has more activity than marketing. Companies like Rox, Monaco, and Actively are pushing up the left side of the AI x GTM Tech Triangle. The AI SDR is seeing a second wave.

Marketing has been slower, but it’s coming.

Within marketing, paid media is the category I see hitting the fastest in B2B. And it’s following the Triangle.

You can buy the agent: a vendor’s agent plans, launches, and optimizes your campaigns. Or you can build the agent and buy the Legos: you own the system, vendors provide infra and tools, and you connect to ad platform APIs and MCPs. Startups and existing players are moving on both paths.

It’s heated up in the last few months.

Profound, the AI search company valued at $1.8B, launched Profound Ads on November 5: “Now it can run your paid ads too.” Primer relaunched on November 3. Different came out of stealth with a $48M seed in October. Demandbase launched Mojo in September. Of the 24 vendors I looked at, 18 made their paid move this year.

On the build side, LangChain open-sourced the agent that runs its paid media. Corey Haines’ marketing skills, with a paid ads skill since January, passed 59K GitHub stars. Meta opened its Ads MCP to every app in July. Startups like Synter and Graphed sell the plumbing your agent needs to run ads.

Tofu started as a content generator for B2B marketers. In October it turned its homepage into “100 ad specialists in one.” It “plans, launches and tunes your LinkedIn, Google and Meta ads from your CRM,” for a flat $2,500 a month. Every campaign is “staged in your own ad account with the campaign switched off, and a person on your team turns it on.”

Different came out of stealth on October 20 with a $48M seed led by General Catalyst. It “launches your email, ads and landing pages,” so paid is one channel of several. “Marketers review and approve all work before it goes live.”

Multiply is an AI-native agency. It left stealth in March with $9.5M and sells outcomes, starting at $6K a month with a strategist.

Petavue is the newest. Prasanna Venkatesan, who co-founded Insent (ZoomInfo bought it in 2021), is building “Agents for B2B paid media that measure, recommend and act on governed metrics.” It’s in early beta for teams spending $30K+ a month. “Petavue proposes; a person on your team approves before anything reaches an ad platform.”

Primer relaunched on November 3 as an assistant that builds and checks your audiences. Campaigns are next: “Over time, we want you to be able to hand Primer more of the work: setting up campaigns, monitoring what’s happening, and helping make adjustments.”

Profound announced Profound Ads on November 5, as a beta you register for. It connects Google, Meta, LinkedIn and ChatGPT, and you “Approve every recommendation before your AI Marketer acts on it.”

Vector is a contact-level ABM platform with one narrow agent. Its Bid Agent “adjusts your LinkedIn bids every 15 minutes.” It “only changes bidding and pacing.”

Metadata is moving toward agents, but the platform is still the product. In February 2025 it launched a LinkedIn Bid Agent in beta, then added analyst and creative agents. Its site now introduces seven named AI “specialists”. They’re a layer over its MCP’s tools, and its own FAQ says the agents aren’t a separate product. Metadata’s bigger move is on the build side.

On the Build side, we’ve seen marketers and GTM engineers sharing experiences and resources in this space for many months. Here are a few that caught my attention.

LangChain’s team open-sourced the agent that runs their paid media: “In six months, it took paid media from 0 to 20% of our marketing pipeline, cut cost per qualified lead about 30% while spend rose 60%, and turned what used to be contractor reports into a two-minute Monday run.” They share the architecture and approach in detail:

Tom Wentworth, incident.io’s CMO, shared a LinkedIn Ads plugin for Claude Cowork in March: “Create campaigns. Clone them. Set audiences. Start/stop. Manage budgets. … Give it an image and tell it to create 5 copy variations and it will.”

Notion is hiring an AI Product Manager, Growth Marketing to build “self-optimizing paid marketing systems” and guide “the agentic infrastructure that brings it to life.”

ColdIQ’s Michel Lieben: “We manage > $300k/mo in ad spend via Claude Code.” In June he gave away the 12 skills they use, from a LinkedIn bid optimizer to a Meta creative fatigue monitor.

Cognition’s Will Hunter has Devin build LinkedIn campaigns: “LinkedIn Ads has no MCP, so we built a wrapped API for Devin to work in safely. Everything is staged as a draft, and going live needs a separate yes.”

Open source is moving fast.

The Legos come in a few flavors. Synter and Cody Schneider’s Graphed are new companies built for agents to run ads (Graphed will also send engineers to build yours). Clay, Primer and Hightouch build audiences and sync them to the ad platforms. Supermetrics and Metadata’s MCP let your agent build and launch campaigns. Vector, Fibbler and Dreamdata have MCPs that show your agent which ads touched pipeline. Eight ad platforms have official MCP servers (but not LinkedIn yet).

If you’re a CMO thinking about building, there are a lot of ingredients available. It’s still going to be significant work to go fully autonomous, but I think you can start to lean on agents for pieces of the workflow.

I’m not technical and I’ve made some progress.

I’m tinkering on the build side myself. Take this with a grain of salt because I’m doing low volumes in paid media.

In a few days, I got audiences syncing and refreshes to run autonomously with my night shift agents. Every night they rebuild my audiences and update them via Primer’s API. Those go into LinkedIn matched audiences and other ad platforms as I expand (Primer handles the matching and syncing account lists to all major channels).

The creative generation part has been fun and interesting. I have skills for Claude Code to compose HTML graphics from my live data. It calls the Codex plugin to generate components as needed (like people), puts all the pieces together in HTML, gives me a review & feedback page, renders my picks to PNG, then loads them into campaign manager. For one campaign in October, asking GTM Tech brands to claim their profile in my Brand Portal, it built 17 ad variants. Seven went to LinkedIn. It also generated a video (but that took a lot more iteration). Claude drafted the seven in LinkedIn Campaign Manager. I reviewed, edited and approved them.

It’s an assistant, not an agent. I’m in every step. I think I could make this more autonomous going forward.

Many parts of marketing are taste-based and will take longer for agents to crack. Paid media feels closer to the other domains where agents are working.

1. It’s relatively verifiable. Not as pure as coding, but you have clearer metrics to optimize vs other parts of marketing. Proving pipeline is still the hard part. In Primer’s customer data, LinkedIn reports 4.4 conversions for each one that confirms as a real lead.

2. It’s a replacement sale. Hire the agent, drop the agency. Very similar to the early AI SDR pitch. The budget already exists as a line item. Metadata’s Instruqt case study is the plainest version: “It makes zero sense to go back to a traditional agency.”

3. LLMs are getting very very good at video edits, graphics, and composing visual assets for B2B ads. I do NOT mean one-shotting image gen in ChatGPT. I mean creating HTML graphics, composing AI generated components, and remixing assets. That’s where my own experiments have paid off most.

It’s the same pattern as the AI SDR’s second wave. Inbound stuck because it looks like support: a bounded job with a clear outcome. Paid looks more like that than the rest of marketing does.

B2C tried this first. In 2017 the lingerie brand Cosabella said it replaced its agency with Albert, an AI: “I would never have a human do this type of work ever again.” Albert’s assets were sold in 2022. There was a wave of AI ad creative startups, mostly for consumer brands. Pencil went to Brandtech in 2023. AdCreative.ai sold to Appier for $38.7M in 2025. Creatify ($15.5M Series A) and Arcads ($16M seed) say they’re near $10M ARR. Meanwhile Google, Meta and TikTok made generative ad creative a free feature. None of it really reached B2B in that early wave.

Now B2B builders and vendors are pushing forward.

The purest form of this is a fully autonomous paid media agent that goal seeks within a budget and set of guidelines. The agent handles all the creative iterations, campaign optimizations, testing, and audiences.

That’s where I expect the most ambitious vendors and builders to go.

Today most vendors aren’t fully autonomous. The agents recommend, but humans approve.

There is clear hesitation from people in the game. Kamil Rextin, whose agency ran LinkedIn and Google end to end with AI: “if the AI makes a mistake you’re completely blind and screwed.”

I’m curious how much you’ll turn over to a paid media agent.

Here’s my bet today. Reply and tell me what you see differently.

It has the right characteristics (somewhat verifiable, increasingly feasible, clear replacement sale). And the blast radius in paid media is smaller than in other parts of marketing (you can’t accidentally spam every customer). It’s a natural place for CMOs to look for agent leverage.

AEO players, ABM incumbents, SaaS Ad-Tech, Clay, startups and open source are all threats to the pure plays. And you have to imagine that Meta, Google, LinkedIn, and other ad platforms will develop native tools.

The evidence on the “build” side has been around longer than the buy side. Getting something basic running (like I did) is pretty fast. Building something fully autonomous is another story, but I think the more sophisticated GTM orgs will try. We’ll see more open-source skills and people building in public. I expect LinkedIn will also have an MCP soon enough, which removes another barrier.

Paid media is following the Triangle. Buy the agent or build your own. Both options have appeal.

This looks like the first B2B marketing category agents crack.

Today it’s more human in the loop (assistants, co-pilots). But the ambitious builders and vendors are heading to full autonomy.

Which vendors and builders should I be tracking? I read all replies.