[JJ’s LLM Insight #4] It’s 6 PM. You Can Close Your Laptop.

Written in

by

Part 3 was about choosing the right model for each job. This time, I want to talk about where the work actually happens. If I’ve handed a task to AI but still can’t close my laptop, something feels off. And if switching computers means explaining everything again, that’s another problem. Both are reasons I’m interested in Kimchi Workspaces.

I Delegated the Work. Why Am I Still Here?

It’s 5:40 PM. I’ve given the AI a task and I’m getting ready to head home. A glance at the screen tells me there are still files to check, changes to make, tests to run, and results to review. Things are moving, but I have no idea whether it’ll take another ten minutes or another hour.

Can I close my laptop? What will survive if I disconnect now? Can I pick it up later? I end up plugging in the charger and waiting a little longer. I started using AI to get work off my plate. Now I’m sitting here watching it work.

I could stop the laptop from sleeping and leave it running. But then the job depends on my laptop staying powered on and connected. Disabling security settings such as the screen lock and walking away isn’t a good answer either. It could violate company policy and leave work files or signed-in accounts exposed to someone else.

If the Model Is in the Cloud, Why Does My Laptop Have to Stay Open?

Using an AI model over the internet doesn’t mean the whole job runs there.

Say the AI decides, “Check this file, change this part, then test it.” That decision may happen in the cloud. But the software that opens the file, makes the change, and runs the test could still be running on my laptop.

That’s what a coding agent does: it takes the model’s instructions and carries out the work. From the chat window, it’s easy to miss the fact that my laptop may still be doing the hands-on part.

Closing the lid or losing the connection can then interrupt the job. A saved conversation doesn’t mean the tests kept running. Being able to reread yesterday’s chat and having the work continue while I’m away are two separate things.

Workspace Is the Workshop. Teleport Is the Way In.

One Kimchi feature I want to highlight is Workspace. Kimchi is part of CAST AI, where I work. Think of a Workspace as a remote workshop equipped with the tools and materials your AI needs. It runs separately from your laptop, and your company can manage the tools and access permissions inside it.

Teleport connects the work you started locally to that workshop. It hands over the conversation and project so the agent can continue in the prepared Workspace.

The flow is straightforward:

Set up an isolated Workspace => Use Teleport to connect and hand over the task => Keep working in the Workspace => Reconnect when you need to

Once the job is running in the Workspace, my laptop becomes the window I use to check progress and give instructions. Closing that window doesn’t have to stop the work. When I reconnect, I can see where things stand, give further direction, or review the results.

A Different MacBook Shouldn’t Mean Starting Over

I have a work MacBook and another MacBook at home. One thing that bothers me is how awkward it can be to continue the same session on the other machine. I spend time explaining a task and agreeing on an approach at work, then have to bring the AI up to speed again at home.

Copying the files sounds like it should solve this, but the files are only part of the story. Why did we start this task? What have we already tried? How far did we get? That’s the context we need to keep going.

Then there are skills and connectors. Skills are the procedures or ways of working I’ve given the AI. Connectors give it access to services such as code repositories and work documents. If they’re set up on one MacBook but missing from the other, I have setup work to do before I can get back to the actual task.

A Workspace gives that working environment a consistent home. Prepare the skills and service connections there, then return to the same Workspace and existing session from whichever device you’re using.

What I want to keepWith the Workspace as the place where I work
ContextContinue from the conversation and progress in the existing session.
SkillsKeep using the procedures and instructions prepared in the remote environment.
ConnectorsUse the work services already connected for use from the Workspace.

I can start on my work MacBook and connect to the same Workspace from my MacBook at home. A browser or supported development tool lets me return to the files and session I was working on. The device changes. The place where the work happens stays the same.

That’s the experience I want: leave the work in the Workspace and change only the device I use to reach it. Opening a different laptop shouldn’t mean starting the job again.

Does My Agent Really Need the Keys to My Laptop?

My laptop holds much more than the task I’m working on: other projects, personal files, and access to several services. When I let an AI edit files and run programs, I need to think about how much of that it should be able to reach. The more work I delegate, the more that matters.

A Workspace gives the agent a separate place to work, with what it needs for the job. The company can decide which tools are installed, which services it can reach, and whose permissions it uses to access work materials. That separation can also help limit the impact when something goes wrong.

Kimchi provides controls for the things a company needs to manage:

What the company managesWhat that means
Working environmentPrepare the team’s tools and settings in a template, so new workspaces can follow the same standard.
External connectionsLimit network access to the services the work requires.
CredentialsManage authentication centrally, with the option to attach credentials to outgoing requests.
Management historySee who created a Workspace or changed its settings.

If the agent needs the company’s code repository and document service, those connections and permissions can be prepared in advance. That reduces the need for developers to copy credentials and repeat the same setup on each laptop.

Whether I connect from my work MacBook or another company-approved device, the work runs in the environment we’ve set up. I can continue my task, and the company can manage where it runs and what it can access. Getting both is a big part of the appeal for me.

Before You Leave, Leave a Decent Brief

Once the workspace is ready, there’s still the question of what to ask the AI to do. “Just take care of it” leaves plenty of room for decisions that may need my input after I’ve gone home.

Take the intermittent logout problem from Part 3. Before heading out, I might leave a request like this:

Find out why users are occasionally being logged out, and prepare a fix. Focus on the login-related code and tests. If the fix needs major changes elsewhere, explain why. Check that the problem no longer occurs after the change. Leave me a summary of the files you changed, the test results, and anything you couldn’t resolve. I’ll review it before we put it into production.

That sets the scope, the checks to run, and the points where a person needs to make a decision. It helps keep a task from growing in unexpected directions while I’m away.

Kimchi’s Ferment breaks longer jobs into stages and keeps track of progress and decisions as the work continues. Workspace provides the place to run the job; Ferment helps organize the steps and define what counts as done. Together, they make longer tasks easier to delegate, with clear points for human review and boundaries for work that can continue.

If I’m still checking the screen all evening, moving the job to a remote machine hasn’t really freed me up. I need to leave it in a state where I can actually step away.

I’m Going Home. The Bill Doesn’t Need to Work Overtime.

This connects to the idea of tokenomics I discussed in Part 3. The cost of using AI includes the time spent explaining things again, redoing work, and reviewing results, as well as the model bill.

I’d assess Workspaces the same way. Being able to close my laptop is useful, but so is being able to resume work easily on another device. Less time rebuilding context, setting up skills and service connections again, or recovering an interrupted task is part of the value.

A remote workspace also has an operating cost. Repeating the same failed attempt all night, or running work nobody needs, can push that cost up. It’s worth looking at both model usage and the cost of keeping the workspace running.

I’d measure the results I can review and actually use. An agent running for hours doesn’t necessarily mean it finished more useful work.

“I Worked All Night” Isn’t the Update I Need

When I open the Workspace the next morning, I want to know what changed, what was checked, and what still needs my decision. That’s what lets me review the result and move on.

Part 3 looked at dividing work between models. This post is about giving that work a place to run and returning to it later. Kimchi Workspace provides an isolated environment, and Teleport connects the task to it. The context and working setup stay there, while I connect when I need to give direction or check the outcome.

At 6 PM, I’d like to close my laptop. And when I open a different MacBook at home, I’d rather not introduce myself to the AI all over again. The work should continue when I move, in an environment my company can manage. That’s the value I see in Kimchi Workspaces.

Tags

Categories

답글 남기기

JJ's technical blog

A blog about the messy, fascinating intersection of Kubernetes and AI – where infrastructure decisions increasingly shape whether AI initiatives actually succeed. Drawing on hands-on experience with enterprise customers across Korea and APAC, this space covers everything from cluster optimization to the evolving demands AI workloads place on cloud-native systems.

Seoul,
South Korea