Every growth channel a startup can reach is now crowded with AI-generated noise. Cold email, programmatic content, LinkedIn outreach, and paid social all feel saturated at once. Raj Singh, VP of Product at Mozilla, has spent his whole career in startups, and he sees one channel still standing. His answer points toward community-led growth marketing, with AI agents doing the heavy lifting behind it.

Raj Singh joined the Season 4 Episode 2 of AMA SF’s Misadventures in Marketing podcast to unpack this shift with hosts Steve Haney and Peter Farago. Their conversation offers a useful frame for modern marketing teams. Humans should own the trust, taste, and strategy, while agents should own the repeatable execution.

Why is community-led growth marketing the last channel that still works?

Singh puts the problem in blunt terms. AI now lets every growth marketer flood every channel at once. In his view, “the only remaining growth channel left is authentic word of mouth referrals.” Buyers no longer know what to trust, so they lean on people they already know.

That shift explains why individuals now carry more weight than the companies they work for. Singh described it as a decade-long trend line. “People follow people, they don’t follow necessarily companies,” he said.

So his advice to founders flips the usual sequence. Pitch him an idea in the food category, and his first question is whether you already have a community in food. If you don’t, he suggests building that audience before writing any code. In his view, acquisition now comes first and the product follows.

Communities also don’t need to look technical. Back in 2009, Singh rallied mobile industry people who rode motorcycles into a group called the Cells Angels. Years later, he started a community of CEOs who practice jiu-jitsu. Any shared interest can become the audience that drives your word of mouth.

How do introverted founders build an audience without performing?

Many engineering founders resist public posting, and Haney sees this hesitation with his early-stage clients constantly. Singh admits he once worried he had nothing good enough to share. Eventually, he stopped caring and simply posted what he thought.

Even so, a public feed is only one option. Singh points to WhatsApp groups, Discord servers, and Reddit threads where quieter founders can act as trusted voices. Farrago added that developers can start by answering questions on Stack Overflow or sharing anonymized customer pain points in blog posts.

Reach also tends to exceed what the numbers suggest. Farrago cited the 90-9-1 rule, a long-standing pattern in online communities. Roughly 90% of members simply lurk, 9% react or reshare, and only 1% create. That silent majority still reads your work, and Singh has felt this firsthand. An old contact invited him to an event because his Threads posts kept appearing in her feed, even though she didn’t follow him.

What does in-person community building look like now?

Haney has reversed a position he held for years. He used to tell clients to skip trade show booths and book back-to-back meeting rooms instead. Now he recommends that early-stage companies, particularly in robotics and physical AI, buy the 10×10 booth and simply show up. He frames it through Paul Graham’s advice that startups should do the unscalable things first.

Singh runs his own version of this discipline. After the Mozilla acquisition, he committed to attending one meetup a week, with a simple exit rule. He leaves as soon as he meets five new people, whether that takes 45 minutes or 90.

Private dinners are another tactic gaining ground. B2B sellers invite peers to network over a meal, and the gathering creates a natural reason to follow up afterward. Singh calls this a relatively low customer acquisition cost for a meeting. Farrago has run these dinners himself, and he notes that the herd has caught on. A few anchor guests now make the difference between a full table and an ignored invite.

How can a small team use AI agents to run the growth playbook?

Community gets you to the first hundred or thousand users. After that, Singh says you still have to run the known playbook, even though every channel feels saturated. AI now lets one person cover ground that once required a multi-person team.

His side project, GPT Food Cam, shows the pattern clearly. Singh built a multi-agent workflow that mirrors a real editorial team. One agent researches, another writes, a third edits, and a fourth fact-checks each superfood article before he approves it. A second agent then turns each blog post into a short TikTok slideshow, and those videos pull in a few hundred views a day. His next step had an agent upload the videos through iPhone mirroring on his laptop, which automates the funnel overnight.

At Mozilla, the same approach scales up. One agent studied competitor sitemaps to shape a small business blog. Another crawled roughly 50,000 podcast transcripts to find small business tips and clip them for TikTok, with a sponsor tag at the end. For Solo, Mozilla’s website builder, a team of agents identified businesses without websites and gathered details from Yelp and Google Maps. The agents then built each business a site in its own brand colors before any outreach went out.

Where should marketers keep humans in the loop?

Singh frames his best ideas around a simple prompt. “Imagine you had 50 interns, what would you do?” The creative leap still belongs to a person, and he believes the craft now lies in translating that thinking into clear steps for an agent.

That means writing instructions with real detail. “Communicate to it like it’s a human,” Singh said, because people tend to give AI one vague line and expect magic. He plans extensively, then has one model build while another reviews, trading roles between Claude and Codex.

His stack is less exotic than people assume. He works mostly in the terminal, and he says a $200 monthly subscription covers everything he builds.

Strategy, however, sits behind a much firmer line. Singh skips competitive monitoring agents and trend dashboards, since they mostly add noise. His agents handle the execution, never the judgment. Automating strategy, he said, would mean “kind of eliminating taste and judgment,” which is exactly where teams differentiate.

What skill will define marketing leadership next?

Singh calls it product editing, which means deciding what not to build or say. When agents can produce almost anything, the hard part becomes choosing the single message a customer can repeat in an elevator. Products that skip this step ship without a soul, and word of mouth has nothing to carry.

That idea brings the conversation full circle. Agents can multiply output, yet only a trusted human voice gives that output a reason to spread. Marketing leaders who build the community first will find their automation has somewhere worth sending people.

Listen to the full podcast on Spotify and Apple Podcasts.

Check out summaries from other episodes:

• Career Positioning in the Strangest Job Market We’ve Ever Seen
• The Marketing Playbook Is Dead. The Job Is the Playbook Now
• GEO, Dark Funnel, Prompt Slop: An AI Marketing Fluency Check

Misadventures in Marketing is a weekly podcast by the AMA San Francisco chapter. Veteran Silicon Valley marketing execs Peter Farago and Steve Haney explore the messy, rewarding, and occasionally absurd world of high-tech marketing  –  especially in early-stage startups. Each episode covers real-world challenges, trends, and lessons from the front lines.