How One Developer Automated Marketing With AI Agents

Brian Casel doesn't want to be a marketer. He wants to build things. But he runs a business, which means marketing happens whether he enjoys it or not. His solution: teach AI agents to handle the parts of marketing that follow patterns.

In a 27-minute walkthrough, Casel demonstrates four custom skills he's built for his AI agents using OpenClaw and Claude. These aren't theoretical workflows—they're production systems running his actual marketing operations for Builder Methods, his AI education business. What's interesting isn't that he's using AI for marketing. It's how he's using it: as infrastructure, not magic.

The Radar Scan: RSS Feeds Meet AI Curation

Every morning at 4 AM, Casel's marketing agent (he named it Veil) runs what he calls a "radar scan." The agent monitors RSS feeds from Twitter searches tracking the Anthropic team, the OpenAI team, the Cursor team, and various AI influencers. It reads through the XML feeds, applies training about Casel's interests, and produces a markdown file summarizing what it thinks he should know.

"These are people and companies and like industry news and influencers and thinkers that I really like to follow," Casel explains. "My agent kind of collects those and analyzes a lot of incoming feeds and then decides which ones the agent thinks based on my training are most relevant for me and my business."

The technical setup is straightforward but layered. RSS.app converts Twitter searches into XML feeds. The agent reads these feeds, cross-references them against training data stored in either custom tools or markdown files, filters for relevance, and outputs a formatted report. It sends a Telegram notification with a link to the full report—never the whole thing, just enough to tell Casel it's ready.

What's notable: this isn't content generation. It's content curation—a different problem that suits current AI capabilities better. The agent isn't creating insights; it's applying consistent judgment about what deserves attention. That's a pattern recognition task, which is exactly what these systems can handle.

Brand Illustrations: When Consistency Matters More Than Novelty

Casel's second skill handles brand illustrations—the visual assets scattered across his website, workshop pages, and social media. Every image follows the same style: specific colors, particular line weights, consistent shadow treatments. The goal isn't artistic innovation; it's brand coherence.

His process combines Claude for concept development and Google's Imagen API for actual image generation (Claude doesn't generate images). When he needs an illustration, he opens a Claude project, describes what he needs, and the agent walks him through an interview: What's the context? Which brand colors? What dimensions?

"Before it gets to work on the actual illustration, Claude is going to use my training and my references and the information that I've provided to generate three potential concepts and it's going to describe those in detail," Casel says. "And that's really important before we get into creating an image."

He picks a concept, the agent documents everything in a project.md file, then sends the refined prompt to Google's API. First draft comes back. Iterate if needed. The agent knows the brand parameters well enough that even V1 attempts land close to usable.

This workflow addresses something specific to visual branding: the problem isn't creating good images, it's creating consistent images. A human designer could do this—and would do it better—but the ROI calculation changes when you need 50 illustrations across various touchpoints. The agent can't match a designer's taste, but it can match its own previous output, which is sometimes more valuable.

The Newsletter Pipeline: From Voice Memo to ConvertKit

Casel's Builder Briefing newsletter goes out weekly. Simple text format, minimal styling, a few sections. He used to create it manually in ConvertKit—formatting, scheduling, all the clicking. Now he doesn't touch the email platform at all.

His workflow: take a 20-minute walk, record a voice memo brain-dumping the main idea he wants to share, feed that to Claude with his newsletter writer skill. The agent drafts the content. Then a second skill handles the actual email setup—formatting it for ConvertKit, scheduling it, pushing it live.

The transcript cuts off before he demonstrates the full process, but the architecture is clear: voice → Claude → draft → review → second agent → published. Each step follows rules he's documented in skill files. The agents aren't writing for him; they're writing with him, handling the parts that follow patterns while he provides the actual thinking.

What This Actually Means

Casel's approach reveals something useful about where AI agents work and where they don't. These skills succeed because they're handling tasks that:

Follow documented patterns
Require consistency more than creativity
Involve multiple tools and APIs
Would otherwise pull a developer away from building

They're not replacing strategic thinking. The radar scan doesn't decide what trends matter—it surfaces options for Casel to evaluate. The illustration system doesn't invent visual concepts—it executes on established brand parameters. The newsletter workflow doesn't generate ideas—it formats and publishes them.

"Most marketing work follows a pattern," Casel notes. "And once you see these patterns, you can turn these processes into skills and hand them to agents."

That framing—"hand them to agents"—is doing interesting work. It positions these tools not as autonomous systems but as delegation targets. You document a repeatable process, encode it as a skill, assign it to an agent. The mental model is management, not magic.

The open question: how many marketing functions actually fit this pattern? Casel has found four. Plenty of marketing involves judgment calls that resist systematization—positioning decisions, messaging pivots, knowing when to double down versus when to move on. Those tasks don't decompose into skills you can hand to an agent. At least not yet.

But for the parts that do follow patterns—the monitoring, the asset generation, the formatting, the distribution—treating AI agents as infrastructure instead of innovation might be the more honest frame. Casel isn't revolutionizing marketing. He's just automated the parts he'd rather not do himself.

Which, for a builder who wants to keep building, might be revolutionary enough.

Dev Kapoor covers open source software, developer communities, and the politics of code for Buzzrag.