Day 6 of 28 · OpenClaw Challenge
Your First Conversation
⏱ 7 min
📊 Beginner
OpenClaw is installed. The API key is configured. Your agent is sitting there, ready, waiting for its first real task.
Today you'll talk to it — and watch the moment it stops being "just AI" and starts being your agent.
Start simple
Don't overthink your first request. Keep it simple. Something the agent can do right now with its built-in tools.
Try this: "What's the weather in my city?"
What's the weather in San Francisco right now?
Right now in San Francisco:
🌤️ 62°F (17°C) — Partly cloudy
💨 Wind: 12 mph from the west
💧 Humidity: 68%
🌅 Sunset at 6:14 PM
Looks like a nice afternoon. Want to know the forecast for the rest of the week?
↻ Replay conversation
That might not seem impressive. Google can do that. But here's the difference: you didn't open a browser. You didn't type a search query. You didn't scroll past ads. You asked your agent, and it went and found the answer for you.
Now try something harder.
Knowledge Check
When OpenClaw fetched the weather, what step of the agent loop was it in when it called the web tool?
The Act step is when the agent executes its plan — in this case, calling the web browser tool to look up the weather. It observed your request, thought about which tool to use, acted by fetching the data, then reported the results.
Now push it further
Ask something that requires the agent to actually work — not just look up a fact:
"Summarize the top 3 tech news stories today."
Summarize the top 3 tech news stories today.
Here are today's top 3 tech stories:
**1. Apple announces new AI features for iOS 19**
Apple revealed on-device AI agents that can manage apps on your behalf. They're using a privacy-first approach — all processing happens on the phone.
**2. EU passes landmark AI regulation bill**
The European Union voted to require AI transparency labels on all consumer-facing AI products. Companies have 18 months to comply.
**3. Open-source AI models close the gap with proprietary ones**
New benchmarks show open-source models matching GPT-4 on 85% of tasks, at a fraction of the cost. The open-source AI movement continues to accelerate.
Want me to dig deeper into any of these?
Tell me more about #3 — which open-source models?
Here's the deeper dive:
The top performers in the new benchmarks:
🥇 **Llama 3** (Meta) — Best overall for general tasks
🥈 **Mistral Large** — Strongest for coding and math
🥉 **Command R+** (Cohere) — Best for RAG and document analysis
Key finding: these models cost 60-90% less to run than GPT-4, and they can run on consumer hardware. The report suggests enterprise adoption of open-source AI doubled in the last 6 months.
↻ Replay conversation
Knowledge Check
What did the agent do differently from a Google search in that example?
B
It searched multiple sources, read the articles, summarized them, and let you drill into details conversationally
D
It used a secret database
A Google search gives you links. You have to click, read, compare, and summarize yourself. The agent did all that work — searched, read, synthesized — and gave you a ready-to-use summary. Then it let you drill deeper with a follow-up question.
The moment it clicks
Here's the thing most people feel during their first real conversation with an agent: "Wait, it actually DID that?"
It didn't give you a template. It didn't link you to a website. It went to the internet, read real pages, pulled real data, and brought it back to you in a format you can use immediately.
This is what the first six days have been building toward. The brain (AI model), the process (agent loop), the tools, the installation — it all comes together in this moment.
Can you create a file on my desktop with a summary of what we talked about today?
Done ✓
Created "OpenClaw_Day6_Summary.md" on your Desktop with:
📄 Weather check results
📰 3 tech news summaries
🔍 Deep dive on open-source AI models
Anything else you'd like me to do?
Happy to help! This is just the beginning — wait until we connect your email and calendar. 🦞
↻ Replay conversation
Final Check
What makes this different from copying ChatGPT's output into a file yourself?
A
The agent's summary is better written
B
The agent created the file directly on your computer — you didn't have to copy, paste, or save anything
C
Nothing — it's the same thing
D
ChatGPT can't write summaries
The agent didn't just generate text — it created a real file on your real computer. No copying, no pasting, no switching between apps. It acted in your world. That's the agent difference.
💬
Day 6 Complete
"You talked. It did. That's not a chatbot — that's an agent working for you."
Tomorrow — Day 7
What Just Happened?
Let's pull back the curtain and read the agent's logs — understanding exactly what it did and why.