If you missed today’s AI Agents Workshop, here’s a full recap of what we covered. From what agents actually are, to how our team at Emarketed is using them every day, to practical tips for getting started yourself.
View the Workshop Slides
What Are AI Agents, Really?
Forget the buzzwords. An AI agent is essentially a smart system that can connect to different tools: the internet, databases, calendars, email, and more. It pulls information back to you in seconds. When you ask it a question or give it a task, it doesn’t just generate a response. It reaches out, gathers what it needs, and delivers an answer.
If you’ve used ChatGPT, you’ve already interacted with an agent. But agents go well beyond chat. They can break a goal into steps, use tools to complete those steps, check their own work, and even hand off to a second agent for quality review. Some tasks take a split second; others, like deep research, might run for 30 minutes before delivering a finished result.
The Agent Landscape
There are a lot of options out there right now. The most popular platforms include ChatGPT, Claude, Crew AI, Autogen, Microsoft Copilot, OpenClaw, and Salesforce Agentforce. Each has its strengths.
Claude is a favorite in the tech and coding community, with models ranging from the highly capable Opus tier down to more affordable options like Haiku for lighter tasks. Crew AI is open source with optional paid cloud hosting. OpenClaw launched at the beginning of this year as an open-source, free platform, and it’s the one we’ve been building on at Emarketed.
Why We Chose OpenClaw
One of the things that drew us to OpenClaw is its flexibility. You don’t need a special interface: it lives inside the messaging apps you already use. Our OpenClaw agents sit right alongside our human team members in Discord, available 24/7.
You can run OpenClaw locally on your own hardware (even a Mac Mini or a phone) or in the cloud. Local is great for anyone concerned about data privacy. You can download and install your own AI model without sending anything to a third-party provider. For cloud hosting, we recommend exe.dev, which comes with its own AI server admin named Shelley. You can literally tell her to install OpenClaw in plain English.
Once it’s running, you connect it to an LLM provider (we use Claude) and a messaging app. That’s really it. OpenClaw can even fix and update itself, which saves a ton of maintenance headaches.
How Emarketed Uses Agents Day-to-Day
Here’s a look at the workflows we’ve built:
Website management: I can message our agent from Telegram and say “change the date on this event” or “update this page.” It connects to GitHub, edits the code, and pushes the change to our Netlify hosting. No developer needed.
Custom graphics: Our agents connect to Google’s Gemini image generation to spin up marketing visuals and social media graphics on demand.
Sales research: One agent runs on a daily schedule, scanning for potential clients based on criteria like business type, employee count, and LinkedIn activity. It even verifies email addresses through Hunter.io so we’re reaching real contacts.
Email outreach: Our main agent has its own email inbox and can send outreach messages on our behalf.
Content creation: An agent researches our industry and writes a daily blog post. It took some coaching early on, but now it’s consistent and reliable.
Social media posting: By connecting Metricool’s API to our agent, we can create and schedule posts across Instagram, Facebook, and LinkedIn, for both the company and my personal accounts, without ever logging in.
Where to Start (and What to Avoid)
The biggest mistake business owners make is trying to automate everything at once. Start with one task. For us, that was the daily blog. Once it was running smoothly, we layered on social media automation, then sales research, and so on.
Think of agents like new employees. They won’t be perfect on day one. You need to monitor the output, correct mistakes, and build trust gradually. Give them a task, get it solid, then add the next one.
Watch Your Costs
A few lessons we learned the hard way:
Instead of giving an agent direct API access, where costs can spiral fast, connect it to a subscription plan. In our first week, an agent got stuck in a loop and burned through $400 in API tokens before the issue was patched. Subscription plans like Claude Max cap your costs and keep things predictable.
OpenRouter is another smart option: deposit a set amount, get an API key, and set spend caps per model. Not every task needs the most powerful model. Use affordable models for research and routine tasks, and save the premium models for coding and complex writing.
Finally, start with limited access. Give your agent only the tools it needs, then expand permissions as you build confidence in its performance.
Wrapping Up
AI agents aren’t science fiction. They’re practical tools that can handle real work today. The key is starting small, staying hands-on, and scaling up thoughtfully.
If you want to see how your business shows up in AI search results, head to Emarketed.com/tools for free tools to check your AI visibility score, scan your site for AI-friendliness, and more.
This post is a recap of the AI Agents Workshop presented by Matt Ramage, founder of Emarketed, a digital marketing agency helping businesses get found on Google, ChatGPT, Perplexity, and AI search since 1998.
