This is a presentation for the Notion MCP Challenge
what i built
OpenClaw just received a blueprint from Amazon Lightsail. No more Mac Minis. No more Raspberry Pis on your desktop acting as your AI agent server. Click deploy and you will have an agent platform running in the cloud.
The AWS samples also have an experimental (non-production) implementation that runs OpenClaw as serverless containers per user in the AgentCore Runtime. Serverless is early, but the direction is clear.
That means OpenClaw can now run in different locations. A Raspberry Pi on my desk. A Lightsail instance in the cloud. Serverless containers on AgentCore or even an EC2. Choose a flavor. (I didn't buy a Mac Mini)
I have 18 agents in mine. These are not toy demonstrations. They solve problems that I got tired of solving by hand.
After re:Invent last year, all the expo vendors in the room started emailing me. Booth scans, follow-ups, drip campaigns. Unsubscribing from each one is death by a thousand clicks. So I built an unsubscribe agent. I do not give you access to my personal mailbox. Sending vendor spam to OpenClaw email inbox. Scan the email, find the unsubscribe link, click on it and confirm. I set up an email rule and forgot about it. 47 supplier lists were deleted in two weeks.
Then there is the train monitor. After rush hour, the next train home is an hour away. If you miss it, you will be standing on a cold platform for 60 minutes. The problem is that trains don't always behave well. Sometimes it arrives a minute early. Sometimes it changes platforms without prior notice. I was updating the train app constantly. The agent polls the live train data and sends me a notification when something changes. Change of platform, early arrival, cancellation. I get the update instead of checking.
OpenClaw even created an entire SaaS-like newsletter platform for me, “The Agentic Engineer.” I wanted a weekly newsletter that would keep me up to date on the Agentic AI content I'm interested in, along with a platform for subscriber management, double opt-ins, click tracking, A/B subject lines, SEO-enabled archive pages, threaded comments, everything. Instead of joining Substack or Beehiiv or whatever, I pointed the request to OpenClaw and let it go. CDK stacks, Lambda functions, DynamoDB tables, SES integration, CloudFront distribution – they structured everything. Then another agent writes and publishes the numbers. The platform works on autopilot. I haven't touched it in weeks. It has more features than most SaaS newsletter tools I've paid for and costs me about $2 a month in AWS bills. An example of true SaaSpocalypse.
Now multiply that by 18 agents, all running on cron programs, and you find the real problem of migrating or cloning your agent work at 10X scale.
The problem of agent migration
Managing 18 agents was already a disaster. SSH into a server. No single view other than the OpenClaw Dashboard. There is no way to pause an agent without editing configuration files or telegramming OpenClaw. There is no complete history of what was executed, what failed, or how many tokens were burned.
But with three deployment goals, a new problem appeared: how do you move your agents between them along with their identity and history?
Each agent has a custom message, personality file, tool configurations, and cron schedules. My unsubscribe bot has email scanning rules. My train monitor has API polling settings. 18 state agents that live in files on disk.
Migrate that from a Raspberry Pi to a Lightsail plane by hand? Copy config files, reedit cron tabs, test each agent one by one? I'd rather stay on that cold train platform for an hour.
I needed a control plane that was portable. Something that could take a snapshot of my entire fleet, move it to a new instance, and recover everything. And I didn't want to run a database for it.
So I built AgentOp
