Скачать или смотреть How to run multiple Clawdbots (aka Openclaw agents)

Presented by Zapier
https://zapier.com/

What happens when you stop hiring people — and instead hire AI agents that manage each other?

In this episode of The Next New Thing, Andrew Warner interviews Bhanu Teja P founder of SiteGPT, who built an autonomous AI “marketing team” made up of specialized agents that plan, communicate, review work, and ship improvements — without human micromanagement.

Bhanu doesn’t run these agents on a fleet of computers. He runs everything on a single VPS. No Mac minis. No local machines. Just a coordinated system of Claude-powered agents that talk to each other through a shared dashboard, assign tasks, create deliverables, review work, and escalate decisions when needed.

SiteGPT is already at $200K ARR, and these agents are actively driving marketing, product improvements, SEO, content, design, and analytics — while Bhanu steps in only when judgment or prioritization is required.

This episode is a rare, concrete look at what agentic workflows actually look like in the real world — beyond demos and hype.

🔗 Guest links

Bhanu Teja P. on LinkedIn: https://www.linkedin.com/in/pbteja199...
SiteGPT: https://sitegpt.ai/


Episode Highlights / Timestamps
00:00 Running 14 AI agents doing marketing autonomously
00:09 SiteGPT reaches $200K ARR
00:18 How AI agents communicate with each other
00:36 Introducing Bhanu Teja P and SiteGPT
00:54 One orchestrator agent delegating all tasks
01:39 Why one chatbot wasn’t enough
01:57 Creating specialist agents by role
02:24 Spawning new agents without coding
02:42 Jarvis: the chief orchestrator agent
03:36 Building a custom dashboard to monitor agents
04:12 Assigning agents to explore the product as real users
04:48 Agents requesting access to bypass Cloudflare
05:24 SEO agent joins automatically
06:00 Agents polling the dashboard every 15 minutes
06:54 Generating a full homepage improvement brief
07:30 Enforcing deliverables for every task
08:15 Turning agent output into production-ready specs
08:42 Handing work to a developer agent
09:09 From recommendation → implementation
09:45 Broadcasting new tasks to all agents
10:30 Agents asking clarifying questions before acting
11:06 Aligning all work to revenue goals
12:36 Letting agents operate independently during vacation
13:03 Preventing agents from publishing prematurely
13:30 Adding a review agent before shipping
14:06 Agent-to-agent “office chat” behavior
14:51 Turning insights into onboarding flows
15:27 The new bottleneck: prioritization, not execution
16:03 Agents deciding what to implement first
16:21 Using analytics to diagnose conversion leaks
17:06 Coordinated SEO, UX, pricing, and dev improvements
17:42 Final walkthrough of SiteGPT


👉 Join us: https://thenextnewthing.ai/