4 AI agents, 1 prompt, 30-day marketing plan in 10 minutes
One prompt. Four agents. A complete 30-day marketing launch plan with 38 pieces of content, generated in about 10 minutes. That is what developer Vivek Shetye reports after building a multi-agent marketing system on top of OpenClaw and wiring it together through Discord.
The test case was a fictional SaaS product called SpotSeeker, a platform for digital nomads to find verified workspaces. The prompt given to the system: “Create a 30-day marketing launch plan for SpotSeeker in New York.” What the system produced is worth unpacking.
The Four-Agent Team
Rather than one generalist agent trying to handle research, strategy, and copywriting at once, Shetye built four specialists:
- Orchestrator Agent: Plans the overall task and assigns work to the other agents.
- TrendScout Agent: Handles research, pulling real-world trend data from the web.
- Growth Agent: Builds SEO and GEO (generative engine optimization) strategy for discoverability.
- Copywriter Agent: Takes the research and strategy output and converts it into finished content.
Each agent runs as an independent entity inside OpenClaw with its own persona, defined responsibilities, and separate short-term and long-term memory. They communicate by mentioning each other inside Discord, just like a real team tagging colleagues in a channel.
How Discord Becomes the Coordination Layer
Each agent maps to a unique Discord bot. OpenClaw routes messages between bots using bindings set in a config file. Channel allowlists keep agents scoped to the right conversations, and agents only respond when mentioned directly.
The result is structured task handoff rather than a noisy free-for-all. Agent A finishes its work, mentions Agent B, and hands off context. No human in the loop to keep things moving.
Shetye also used Discord threads for each task rather than a single shared channel. This keeps context clean per task and avoids token bloat from unrelated conversation history bleeding into each agent’s working memory.
️ Skills the Agents Built Themselves
Both the TrendScout and Growth agents dynamically created skills during the run to handle specific tool needs:
- Fetching data from Google Trends
- Handling rate limit errors (HTTP 429 responses)
- Pulling web insights via SearXNG
These skills let the agents identify trending cities, extract keyword demand, and build the SEO strategy without pre-programmed tooling for every scenario.
The Model: Minimax M2.7
Shetye chose Minimax M2.7 as the underlying model for all four agents. He reports it performs well for multi-step reasoning, agent coordination, and long workflows. The system’s reliability during the run is attributed in part to that model choice.
What Came Out After 10 Minutes
The final output from a single prompt against a fictional NYC launch:
- 15 LinkedIn posts
- 20 Twitter threads
- 3 YouTube video scripts
All 38 pieces were described as data-backed, SEO-optimized, and on-brand for the SpotSeeker product. The full setup, prompts, and configs are available on GitHub via the video description.
Three Design Lessons Worth Stealing
Whether or not you use OpenClaw specifically, the structural decisions Shetye documents are transferable to any multi-agent build.
Specialists beat generalists
A single LLM juggling research, strategy, and copywriting in one context produces lower quality output than four agents each owning one domain. The cognitive load is smaller, the instructions are tighter, and the results are more focused.
Memory compounds over time
Because each agent maintains its own memory, performance on repeated tasks improves as the system accumulates context. The first run is rough. Subsequent runs with feedback get better. That is the same dynamic you would expect from a human hire.
Autonomy needs structure
Without clearly defined output formats, responsibilities, and instructions, agents drift or loop. Shetye’s framing: give agents freedom, but with structure. The Discord thread approach and strict agent-to-agent mention protocol are practical implementations of that principle.
The Catch
This is a developer-built system, not a packaged product. Running it requires setting up OpenClaw, configuring Discord bots, wiring bindings in a config file, and understanding enough about agent architecture to define four coherent personas with separate memory. The full setup is on GitHub, but it is not a one-click install.
If you have the technical tolerance for that setup, the output numbers are hard to ignore: 38 pieces of structured content in 10 minutes from a single natural language prompt.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
