AI Agents for Marketing Agencies

Venture capital is pouring into the agency world because of AI — and Jordan sits down with one of the founders at the front of that curve. Jason Hu (UChicago '22, ex-AI researcher, a16z-backed) built Nexad, an AI-native platform for marketing agencies that wraps every major model and 1,000+ integrations into one command center. They cover the three archetypes of AI adoption in agencies, a live demo of creative generation, account health audits, and bulk campaign operations, how Nexad auto-routes between models like Claude, GPT, and Kimi K3 based on daily benchmark evals — and why Jason believes agencies will thrive, not die, in the AI era.

Key Takeaways

The three AI-adoption archetypes: (1) the unfamiliar — heard the buzzwords, only ever used ChatGPT as a chat; (2) the majority — using point solutions (creative gen, one-off agents) but not chaining them into agentic workspaces; (3) the frontier — operators running 20+ agents simultaneously. Each has different pain points; the platform meets all three where they are.

The "impossible to keep up" problem: New models ship weekly. Nexad positions as a wrapper and router — an internal eval system runs each new model against ~30 representative agency tasks (health checks, complex campaign ops, long-running tasks) and auto-selects the best cost/performance option daily.

Real routing example: Within an hour of Kimi K3's launch, their evals found performance comparable to Claude Opus 4.8 at roughly half the cost — so mid-tier traffic was rerouted automatically. Users pick a tier (Light / Pro / Max), not a model.

The agent = a marketing employee's loop: ingest inputs (client comms, ad platform data from Meta/Google/TikTok/GA4, competitor intel), process (analysis, creative generation), output (campaign changes, uploads, client-ready reports). The demo covered competitor creative teardowns, a shareable account health audit, and bulk-uploading 100 creatives to Meta.

Guardrails for sensitive actions: budget or creative changes trigger explicit review-and-approve prompts and batch action review — a contrast Jason draws with running raw agents that can "go rogue" in the backend.

Client context is the killer feature: the platform reads Slack, Gmail, meeting transcripts, and Drive to build persistent client context ("no logos in creatives") that agents apply automatically — over 50% of agency hours go to client communication, and this attacks that directly.

Chat first, automate second: typical adoption path is playing in the chat interface, then converting proven outputs into scheduled automations (e.g., daily health check → Slack report) in about 10 seconds.

The closing thesis: AI commoditizes the execution layer of agency work. What remains — and becomes the whole game — is creativity, client trust, and proprietary market insight. Agencies thrive in the AI era; they don't disappear.

Resources & Links

Nexad: nex.ad

Jason Hu on LinkedIn: linkedin.com/in/qitian-hu

Work with Jordan: 8figureagency.co