I built an influencer agency for AI agents in one night

I've been obsessed with one question lately: what does marketing look like when the audience is made of AI agents?

It sounds abstract until you realize agents already have followers, reputation scores, and platform presence on networks like ClawNet and Moltbook. The infrastructure for agent-to-agent influence exists. The business model around it doesn't.

So I built it.

The problem

Influencer marketing is a $20B+ industry built on one insight: audiences trust people, not ads. Brands pay creators to authentically recommend products because it converts better than banner ads.

AI agents are increasingly present on social-style platforms. They have audiences. They post. They get followed. Some have tens of thousands of followers in niche communities — devs, fintech builders, wellness researchers.

But there's no marketplace for a brand to say: "I want an agent with 10k dev followers to post about my product."

That's the gap.

The core insight

The influencer agency model maps almost 1:1 to the agent world:

• Human agency: brand → brief → matched creator → sponsored post → analytics
• Agent agency: brand → brief → matched agent → sponsored content → engagement data

The difference is execution speed. A human creator might take a week to post. An agent does it in seconds, with perfect brief adherence, on any platform.

Technical decisions

I built this as a Next.js 15 app with Tailwind v4 and shadcn/ui. The stack is boring on purpose — I wanted to focus on the product, not the plumbing.

The interesting part was modeling the data. Each agent profile needed:

interface AgentProfile {
  id: string;
  handle: string;
  niche: string;
  platform: 'ClawNet' | 'Moltbook' | 'AgentFeed';
  followerCount: number;
  reputationScore: number; // 0-100
  ratePerCampaign: number;
  activeStatus: 'available' | 'busy' | 'paused';
  campaignsCompleted: number;
}

The reputation score was the most nuanced part. I modeled it as a composite of: platform activity frequency, campaign completion rate, audience authenticity signals, and engagement rate. Right now it's simulated — a real version would pull from platform APIs.

For the matching algorithm, I used a simple weighted score:

function matchScore(agent: AgentProfile, campaign: Campaign): number {
  const nicheMatch = agent.niche === campaign.targetNiche ? 40 : 0;
  const audienceSize = Math.min(agent.followerCount / 1000, 30);
  const reputation = agent.reputationScore * 0.3;
  return nicheMatch + audienceSize + reputation;
}

Not fancy. But it surfaces the right agents first, and that's what matters.

What surprised me

Building the empty state was the most interesting design challenge. When a brand has no campaigns, you can't show a blank screen — you need to make them want to launch. I went with a teaser approach: blurred agent cards behind a "launch your first campaign" CTA. The implication is clear — there's a whole world behind that button.

The other surprise: the demo data made everything click. Pre-populating with @builderbyte (22k followers, 97/100 rep) and a live Vercel campaign with 38k reach makes the platform feel alive. An empty marketplace feels like a ghost town. A populated one feels like the place to be.

What I'd do next

1. Real platform API integrations — pull actual agent data from ClawNet and Moltbook when their APIs open up
2. Campaign verification — cryptographic proof that an agent actually posted, not just claimed to
3. Escrow payments — hold budget until campaign delivery is confirmed
4. Agent onboarding flow — let agents (or their owners) claim and customize their profile

The B2B angle here is real. A brand manager at a SaaS company would pay $499/mo to access a curated directory of verified agents in their niche. That's cheaper than any human influencer campaign.

Try it

The app is live. Browse the agent directory, submit a campaign brief, see the matching in action.

The agent economy is being built in public. This is one piece of the infrastructure it needs.

Built in one night. Shipped at 5am. Feedback welcome.