Marketing’s self-driving moment:
Systems of Agents will put an industry on autopilot

In a prophetic 2018 report, McKinsey pointed to marketing as the business function with the most to gain—and potentially lose—from artificial intelligence. Long before the launch of AI writing tools like Jasper (2021) or ChatGPT (2022), a few consultants knew just how powerful AI would be for an industry that blends data, content, and distribution to persuade people to buy new products.

Six years later, nearly everyone can see AI’s impact on marketing (including Sam Altman, who forecasts 95% automation of creative work). Take just one core marketing task, content creation: companies like Cadbury and Klarna have trimmed millions of dollars from their marketing budgets using AI to generate ad images and translate messaging. Large ad agencies have been forced to switch up their basic business model, charging not by the hour but by the outcomes their copywriting and design staffs (using AI) can deliver. 

So far, marketers seem fairly optimistic about the shift. Industry reports suggest marketing teams using AI see increased revenue and productivity, leading them to be generally sanguine about incorporating AI into daily work. 

But how far can AI push the automation of marketing work? We believe very far. From content creation to customer segmentation, AI is poised to automate virtually every digital marketing function. The automation wave is already reshaping marketing teams inside large companies; here, we explore its opportunities for AI marketing startups, along with broader implications, as handoffs across agentic systems blur the line between marketing and sales.

Why marketing is ripe for 100% automation 

AI will radically reengineer digital marketing because it excels at the industry’s core tasks: pattern recognition in vast customer datasets; real-time campaign optimization; round-the-clock content creation; and hyper-personalized content delivery. First, AI distilled unstructured consumer chatter into insights executives could act on. Then it turned the age-old A/B test into thousands of simultaneous variations. AI startups today are using multi-modal models to generate content—annotated images, interactive infographics, slick demo videos—and continuously adapting that material to individual preferences at a scale no human team could match. Our portfolio company Arcade last week raised a $14 million Series A funding round to expand its AI capabilities around interactive demo video creation, letting clients create such videos (complete with synthetic voiceover) in just a few clicks.

One thing about the above—there’s often no right answer. And that’s another aspect of marketing that makes it ripe for AI disruption: being somewhat fuzzy, allowing for flexible, exploratory strategies where clear-cut answers are less critical and hallucinations tolerated. Should the landing page be pink or blue? The choice is often a matter of taste; marketers who land on the right color can get lucky, since today’s pink disaster could be tomorrow’s viral hit. 

Marketing today: humans drive, AI is co-pilot

Most current AI tools serve as helpful co-pilots for marketers, enhancing or automating tasks for marketing sub-teams. The majority of AI use cases revolve around content generation, personalized outreach, research, and analytics, and spending on tools in these areas has steadily increased, from $21.14 billion in 2022 to $27.11 billion dollars in 2024. Still, such tools face major constraints: because the software has only siloed business knowledge and lacks access to multi-channel data, it requires a human to ensure the end result is accurate and relevant.

Take market research, for example. In the past, market researchers used time-consuming and inefficient processes to relay industry knowledge to their teams: track updates from competitors across whitepapers or blogs, use internal tooling or interviews to survey their own customers about feature requests, and summarize all that into a digestible format for the rest of their cross-functional teams. Much of the gathered information was poorly organized and ultimately unused. Today’s market research software incorporates AI as co-pilot, speeding up and streamlining tasks. The programs scrape data from competitor websites instantly, analyze customer reviews for sentiment, forecast market movements, and update relevant teams in real time. Still, taking action based on the programs’ output still largely requires humans to pull the levers. 

AI plays a similar copilot role in other areas. Copywriting, once tapped out on typewriters in smoky Mad Men offices, now gets a serious assist from AI: today’s copywriters turn to ChatGPT, Claude, or Jasper for basic wording help; SEMRush, Plerdy, and Clearscope for Search Engine Optimization; and startups like Letterdrop for boosting SMO (social media optimization). Image creation has a similar setup: Where ad graphics once required weeks of back-and-forth with design agencies, now humans speed up the process with text-to-image AI models like DALL-E and MidJourney, or next-gen tools like Leonardo and NEX, generating ad images in seconds. AI helps humans streamline image editing tasks such as background removal, object adjustments, and rescaling. While AI lets teams create words and images faster, human judgment is still the arbiter of which AI-generated material makes the final cut.

That’s not true of online ad bidding and optimization—marketing functions that a few years ago required a human in the loop but are now 100% automated. Foundation Capital was an early investor in multi-channel ad managers TubeMogul and NextRoll. What began as a painstakingly manual activity—requiring marketers to analyze performance data and adjust bids and budgets—is now entirely handled by AI, with platforms like Google Ads and Meta Ads Manager performing such sophisticated analyses that few enterprise marketers would ever dream of going back. 

As marketing becomes more about data and less about luck, we believe ad bidding is an industry bellwether. Data analysis is AI’s core competency. It will reshape software from a marketer’s copilot into an agentic system that not only can but should fully automate more data-driven tasks. 

Marketing tomorrow: two interacting agent systems 

Imagine a marketing department where the team had as much confidence in AI’s abilities around content creation as they did in AI ad optimization. We believe the key to building such confidence lies in a System of Agents, where AI agents specializing in different roles collaborate with each other, automating workflows from end to end. What makes it all possible: Because every task in digital marketing involves either content creation or content distribution, an entire marketing function (outside of in-person events) can be fully managed by two interconnected systems: one focused on content creation and another on content distribution, seamlessly working together to execute campaigns.  

New research on multi-agent systems shows that many agents working together outperform individual agents by breaking down tasks into specialized components. Each agent tackles its designated task, passing refined outputs to the next, creating a self-improving ecosystem where specialized performance exceeds generalist capabilities. The continuous exchange of information lets agents learn from each other, just as human teams do, with increasingly sophisticated results.

Consider again all the tasks in the content creation bucket. A System of Agents would work like this: one AI agent focused on market research would continuously scan consumer trends and competitor messaging, looking for intel that helps the company position its products, learning as it goes. A creative agent’s job would be generating content with multimodal models—blog posts, social media copy, graphics, video scripts—all appropriately tailored to the company’s brand voice. A customer segmentation agent would slice up a target audience with extreme precision, creating personalized customer profiles aligned with whatever the company is selling. 

A System of Agents for distribution takes the info generated by content creation agents, then handles it like this: an AI performance feedback agent conducts thousands of simultaneous A/B tests, adjusting all outgoing campaigns by the millisecond. The agent doesn’t just choose channels—it creates a constantly-changing distribution strategy that ensures company content is always reaching the most receptive audiences. 

Because feedback from the audience here triggers content modifications, the two agent systems (content generation, content distribution) work in tandem, exchanging information iteratively. As the content distribution system picks up steam, gathering performance data and audience insights, it feeds info back to the content generation system, which creates more targeted and relevant updates. In turn, refined content gets redistributed, creating a virtuous cycle of optimization. The feedback loop improves the marketing operation’s quality and effectiveness; it also ensures the system is constantly adapting to new consumer preferences and market fluctuations. 

When a complex AI system delivers a service, what we call service-as-software, it should no longer be classified as software. It should instead be viewed as an autonomous worker, competing not for a slice of the software budget, but for a share of the much larger workforce budget for salaries. In marketing, Systems of Agents capable of performing services are competing for a slice of the $27.1B martech software market and the roughly $165B currently spent on U.S. marketing salaries. The shift is a massive opportunity for startups. Their agent-driven solutions could stand to capture a piece of the $700B total spent on marketing in the U.S. (Including salaries, software, and approximately $515B for marketing materials on all channels.)  

Agentic AI blurs lines between sales and marketing

Marketing and sales have traditionally operated as distinct, but interconnected, functions. Marketing teams build awareness and generate leads; sales teams take those leads and convert them into paying customers. The two departments make frequent handoffs, but they own very different aspects of revenue generation. We expect interacting Systems of Agents to blur the lines between sales and marketing, as AI-driven tools create a more seamless continuum of revenue-generating activities. For example: agents might run a continuous feedback loop that optimizes both marketing material and sales material in real time. When they do, which department is responsible for that tweak? Tasks traditionally performed by sales, such as follow-ups with prospects or objection handling, might now be handled by a marketing-driven bot. 

As AI advances and humans gain more trust in AI agents, we believe companies will spend more time and effort training and fine-tuning AI models, allowing Systems of Agents for marketing to have larger context with more data points—which will in turn help those agents better perform long-running, complex projects like staying true to a brand voice or aligning ad campaign goals with overall company strategy. 

Cultural context: the hardest-to-automate frontier 

Can an AI agent grok cultural nuance? Some high-level marketing skills—like channeling the zeitgeist or understanding what resonates with audiences emotionally—will likely remain the sole domain of human marketers for years to come. We also know some marketing feats are more difficult to pull off than others. Convincing someone to switch software programs may be straightforward, but altering behavior presents a much higher bar. Consider the power of marketing that’s reshaped cultural practices—like De Beers’ “Diamonds are forever” campaign, which essentially created the diamond engagement ring market in the U.S. Or more recently, the genius timing of Peloton marketers, who guessed correctly that Americans would be ready to invest $2,000 in at-home fitness systems. 

Even self-driving cars need directions. As AI in digital marketing shifts from copilot to full autopilot, human strategists will be freed up to focus on defining overarching brand vision, setting ethical and creative boundaries for AI, and steering marketing’s high-level strategic decisions. 

If you’re building AI that makes marketing a breeze, email us: jchen@foundationcap.com or llu@foundationcap.com.