Try OpenClaw for sending emails and booking a flight
I gave OpenClaw three hours of my life. Task: I gave OpenClaw three hours of my life. It gave me two emails and an existential crisis. Let me tell you about my weekend. I sat down with OpenClaw — the open-source AI agent that’s been lighting up tech socials like it’s the second coming of Christ. The pitch is seductive: an autonomous agent that lives on your machine, browses the web, reads your files, manages your workflows. A digital butler with root access and ambition. So I gave it a spin. Three hours later, I was $15 poorer, over a million tokens deep, and the proud owner of... two sent emails. Two. That’s $7.50 per email. I could’ve hired a carrier pigeon. I could’ve walked to the recipient’s house. Oh, and I also tried to book a Ryanair flight. You know, the thing humans do in roughly four minutes while half-asleep on the couch. OpenClaw couldn’t connect to its own built-in browser. So that obviously didn’t work. The most budget airline in Europe defeated the most hyped AI agent on the internet. There’s a poetry to that. The Slop Machine And it’s not just me having a rough time. OpenClaw has already earned itself a reputation in the open-source community — and not the good kind. Jeff Geerling, one of the most respected voices in open-source infrastructure, recently called out how AI agents like OpenClaw are being used to generate mass pull requests — floods of low-quality, AI-generated code changes submitted to open-source projects. The kind of contributions that look productive on a dashboard but create real work for maintainers who now have to review, reject, and clean up the mess. It’s not contribution. It’s pollution. So let’s recap: it burns through tokens like a bonfire, while literally drinking tons of water, it can’t book a flight, it’s being used to spam open-source repos with AI slop, and — oh yes — OpenAI just acquired it and hired its original creator. Fantastic. Let’s all jump on the bandwagon. ...That’s What I Would Say That’s what I would say, if I couldn’t see what this thing is actually pointing toward. Because here’s the uncomfortable truth: OpenClaw isn’t the product. It’s the prototype. And prototypes are supposed to be terrible. That’s the whole point. Strip away the token hemorrhage and the broken browser, and what you’re actually looking at is the first serious attempt to run an AI agent autonomously on someone’s personal machine — with access to the filesystem, the browser, passwords, and full context about the user’s environment. Not a chatbot in a browser tab. Not a fancy autocomplete. A thing that acts on your behalf, in your world, with your tools. That’s new. And that matters. The Acquisition That Tells You Everything OpenAI didn’t buy OpenClaw because it sends great emails. They bought it because it’s the missing limb. Think about it. ChatGPT is a brain in a jar. It can think, reason, write, and code — but it can’t do anything. It can’t open your laptop, check your calendar, browse a website, or click “Book Now.” It lives behind a text box, waiting for you to copy-paste the world into it. OpenClaw is the body. And now consider who owns 27% of OpenAI: Microsoft. The same Microsoft that makes Windows, Outlook, Edge, Teams, and — oh right — Copilot. The same Copilot that’s already embedded in Office, in your IDE, in your operating system. Connect the dots. An autonomous agent framework, plugged into the world’s most powerful language model, backed by the company that owns the operating system on 72% of the world’s desktops. This isn’t a product announcement. It’s an infrastructure play. The Future Is a Phone Call Away (That You’ll Never Make) Here’s where it gets real. ElevenLabs — the voice AI company — was already integrated into OpenClaw. And we already know OpenAI’s own voice capabilities are eerily good. Combine an autonomous agent with realistic voice synthesis, and you’re not just looking at an AI that books flights. You’re looking at an AI that calls the restaurant and makes a reservation. In your voice. Or a better one. I think we’re about a year away from this: You tell your agent — Copilot, ChatGPT, whatever it’ll be called by then — “Book me a flight to Thailand two months from now. Find a hotel near the beach. Nothing over $120 a night. Oh, and schedule a massage for the second day.”; tools: OpenClaw. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include local machine access and potential integration with voice AI (ElevenLabs). Constraints involve high operational costs, unreliable browser connections, and risk of generating low-quality code or spam. The source reliability is secondary, and human approval is mandatory before sending messages or bookings to prevent errors..
Tools
Published on Agent Work Atlas:
Overview
I gave OpenClaw three hours of my life.
Workflow
I gave OpenClaw three hours of my life. It gave me two emails and an existential crisis. Let me tell you about my weekend. I sat down with OpenClaw — the open-source AI agent that’s been lighting up tech socials like it’s the second coming of Christ. The pitch is seductive: an autonomous agent that lives on your machine, browses the web, reads your files, manages your workflows. A digital butler with root access and ambition. So I gave it a spin. Three hours later, I was $15 poorer, over a million tokens deep, and the proud owner of... two sent emails. Two. That’s $7.50 per email. I could’ve hired a carrier pigeon. I could’ve walked to the recipient’s house. Oh, and I also tried to book a Ryanair flight. You know, the thing humans do in roughly four minutes while half-asleep on the couch. OpenClaw couldn’t connect to its own built-in browser. So that obviously didn’t work. The most budget airline in Europe defeated the most hyped AI agent on the internet. There’s a poetry to that. The Slop Machine And it’s not just me having a rough time. OpenClaw has already earned itself a reputation in the open-source community — and not the good kind. Jeff Geerling, one of the most respected voices in open-source infrastructure, recently called out how AI agents like OpenClaw are being used to generate mass pull requests — floods of low-quality, AI-generated code changes submitted to open-source projects. The kind of contributions that look productive on a dashboard but create real work for maintainers who now have to review, reject, and clean up the mess. It’s not contribution. It’s pollution. So let’s recap: it burns through tokens like a bonfire, while literally drinking tons of water, it can’t book a flight, it’s being used to spam open-source repos with AI slop, and — oh yes — OpenAI just acquired it and hired its original creator. Fantastic. Let’s all jump on the bandwagon. ...That’s What I Would Say That’s what I would say, if I couldn’t see what this thing is actually pointing toward. Because here’s the uncomfortable truth: OpenClaw isn’t the product. It’s the prototype. And prototypes are supposed to be terrible. That’s the whole point. Strip away the token hemorrhage and the broken browser, and what you’re actually looking at is the first serious attempt to run an AI agent autonomously on someone’s personal machine — with access to the filesystem, the browser, passwords, and full context about the user’s environment. Not a chatbot in a browser tab. Not a fancy autocomplete. A thing that acts on your behalf, in your world, with your tools. That’s new. And that matters. The Acquisition That Tells You Everything OpenAI didn’t buy OpenClaw because it sends great emails. They bought it because it’s the missing limb. Think about it. ChatGPT is a brain in a jar. It can think, reason, write, and code — but it can’t do anything. It can’t open your laptop, check your calendar, browse a website, or click “Book Now.” It lives behind a text box, waiting for you to copy-paste the world into it. OpenClaw is the body. And now consider who owns 27% of OpenAI: Microsoft. The same Microsoft that makes Windows, Outlook, Edge, Teams, and — oh right — Copilot. The same Copilot that’s already embedded in Office, in your IDE, in your operating system. Connect the dots. An autonomous agent framework, plugged into the world’s most powerful language model, backed by the company that owns the operating system on 72% of the world’s desktops. This isn’t a product announcement. It’s an infrastructure play. The Future Is a Phone Call Away (That You’ll Never Make) Here’s where it gets real. ElevenLabs — the voice AI company — was already integrated into OpenClaw. And we already know OpenAI’s own voice capabilities are eerily good. Combine an autonomous agent with realistic voice synthesis, and you’re not just looking at an AI that books flights. You’re looking at an AI that calls the restaurant and makes a reservation. In your voice. Or a better one. I think we’re about a year away from this: You tell your agent — Copilot, ChatGPT, whatever it’ll be called by then — “Book me a flight to Thailand two months from now. Find a hotel near the beach. Nothing over $120 a night. Oh, and schedule a massage for the second day.”
Reported result
Actual outcomes reported include high token consumption (over 1 million tokens) and significant cost ($15) for minimal output (two emails). The agent failed to book a Ryanair flight due to browser connectivity issues. No successful complex task completion is evidenced; the tool is described as a prototype with poor reliability.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include local machine access and potential integration with voice AI (ElevenLabs). Constraints involve high operational costs, unreliable browser connections, and risk of generating low-quality code or spam. The source reliability is secondary, and human approval is mandatory before sending messages or bookings to prevent errors.
Try it
I want to test OpenClaw for automating my email and flight booking tasks. Please set up a workflow where you draft two emails to [replaceable recipient emails] and attempt to book a flight from [replaceable origin] to [replaceable destination] on [replaceable date]. Before sending any emails or confirming any bookings, you must pause and show me the draft content and booking details for my explicit approval. Monitor and report the token usage and estimated cost for this session. If the browser connection fails, log the error and stop the flight booking attempt immediately. Do not proceed with any actions without my confirmation. Start with a small sample task to verify connectivity and cost efficiency before scaling up.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.