For a stretch, I left my laptop open overnight so a Claude Code enrichment job could grind through a target list by morning. Lid up, screen dimmed, charger in, like leaving the office lights on so the work could keep happening without me. It worked, and it also felt slightly absurd, my agent's productivity was hostage to whether my laptop went to sleep.
That is the difference cloud execution makes. Claude Code is the same agent either way; what changes is where it runs. Local when you are hands-on and iterating. Cloud when the job should run unattended, on a schedule, without your machine as a dependency. This guide is how to choose, and the tradeoffs on each side.
Same agent, two places it can run
The first thing to get straight: cloud versus local is not a quality decision. It is the same Claude Code, the same capabilities, the same setup traveling with you. What differs is where the work physically happens, on your machine or on managed infrastructure, and that difference maps cleanly onto what kind of job you are doing.
So the question is never which is better. It is where do I need to be for this job. Sitting with it, steering each step, or handing it off to run while you do something else. Answer that, and the choice makes itself.
- You are at your machine, iterating hands-onLocal
- It should run on a schedule, unattendedCloud (Routines)
- Kick it off from your phone, check it laterCloud / remote
Local: hands-on, your files, your machine
Run locally when you are in the loop: iterating on a prompt, working through a messy CSV, building a skill, watching each step and course-correcting. Local execution is immediate and tactile, the files are right there, the feedback is instant, and you are part of every turn. This is most of the day-to-day, deep work at your desk.
Local is also the natural home when the data should not leave your machine at all, or when you are moving fast enough that any round trip would just slow you down. If you are actively driving, local is where you want to be.
Cloud: unattended, scheduled, remote
Run in the cloud when nobody needs to be watching. A long enrichment across a thousand accounts. A pipeline-hygiene pass that runs every Monday at six. A job you start and want to finish whether or not your laptop is awake. Cloud execution takes your machine out of the dependency chain, which is exactly what unattended work needs.
This is also where scheduled work lives. A routine that runs on a cadence belongs in the cloud by definition, because the whole point is that it happens without you initiating it. Local is for when you are there; cloud is for when you are not.
TipIf a job should survive your laptop going to sleep, it belongs in the cloud. That single question sorts most work into the right place.
The data question
Where the work runs changes where the data goes, and for GTM that is worth a real thought, because you are handling customer records, pipeline, and sometimes sensitive account detail. Local keeps data on your machine and inside the tools you connect. Cloud execution means the work happens on managed infrastructure, so it is worth knowing your organization's stance before you push sensitive data through it.
This is not a reason to avoid the cloud; it is a reason to be deliberate. Match the sensitivity of the data to the place you run the job, and when in doubt, check your company's policy. The right answer is usually clear once you name what data the job actually touches.
TipBefore sending sensitive customer data through a cloud run, know your org's policy. Match the data's sensitivity to where the job runs, deliberately, not by default.
Remote control: start it from anywhere
The two worlds meet in remote control. You can kick off or check on a run from your phone, so a job you started at your desk is one you can nudge from a hallway, and a cloud run is one you can approve from the elevator. The mobile app becomes the remote for work happening elsewhere.
For a GTM operator whose day is a corridor of meetings, this is the quiet superpower: the dead ten minutes before a call becomes the ten minutes you unblock the agent, check the overnight job, or approve the draft it prepared. The work does not wait for you to be back at your desk.
The habit that makes remote control pay off is leaving a job in a checkable state before you step away. A quick note to the agent about what done looks like, and permissions set so it pauses rather than guesses at the risky step, means that when you glance at it from your phone between meetings you get a clear yes-or-no to approve rather than a tangle to decipher. Remote control is only as good as how approvable you made the run.
Cost and speed tradeoffs
The tradeoffs are practical. Local has no round trip, so hands-on iteration feels instant, and it uses your own machine's resources. Cloud runs without tying up your laptop and handles long or scheduled jobs gracefully, at the cost of running on managed infrastructure you are billed for and reasoning about.
In practice most operators land on a blend: local for the interactive work where speed and control matter, cloud for the unattended and scheduled runs where not-babysitting-it matters. You are not picking a side once; you are picking per job, and the same workspace supports both.
- Local: instant iteration, data stays on your machine, uses your resources, needs you present.
- Cloud: unattended and scheduled runs, no laptop dependency, managed infrastructure, runs without you.
- Blend: local for hands-on work, cloud for long or recurring jobs. Choose per task.
A quick decision checklist
When you are not sure where a job belongs, three questions settle it. First, do you need to be in the loop turn by turn. If yes, local, because hands-on iteration wants the immediacy of your own machine. Second, should it survive your laptop going to sleep. If yes, cloud, because a job that depends on your lid staying up is a job in the wrong place. Third, does it run on a schedule. If yes, cloud by definition, because the whole point is that it happens without you starting it.
Layer one more question on top for anything touching sensitive data: does this fit our policy for where that data can go. Answer that honestly before convenience decides for you. Run those four questions and almost every job sorts itself cleanly, which is the point, cloud versus local should be a five-second decision, not a standing debate.
- In the loop turn by turn? Local.
- Should survive your laptop sleeping? Cloud.
- Runs on a schedule? Cloud.
- Touches sensitive data? Check the policy before you choose.
Where it goes wrong
The big one is running unattended without guardrails. A local job you are watching is safe because you catch a wrong turn; a cloud job running at three in the morning has no one at the wheel. Before you let anything run unattended, set the permissions, put validation in place, and add an alert on finish, so the job has a fence and reports in.
The other is a data assumption made without checking. Pushing sensitive customer data through a cloud run because it was convenient, without knowing whether that fits your org's policy, is the kind of shortcut that is fine until it is a problem. Name the data, know the policy, then choose.
- Letting a job run unattended with no permissions, validation, or completion alert.
- Assuming cloud is fine for sensitive data without checking your org's stance.
- Using cloud for tight, hands-on iteration where local would be faster and simpler.
- Leaving your laptop open for a job that should have been a cloud run in the first place.
The GTM version
Here is how it settles in a real week. The hands-on work, tuning a skill, working a messy list, building a brief, happens locally at my desk, where iteration is instant. The long enrichment and the Monday pipeline pass run in the cloud, guarded by permissions and a validation check, with a Slack ping when they finish. And I start and approve both from my phone between meetings.
The open-laptop tax is gone. Nothing important depends on my lid staying up, and the work that should run without me does. Which of your recurring jobs is still hostage to your laptop being awake?
How to set it up
Do hands-on work locally
For interactive work where you are iterating turn by turn, run Claude Code on your machine. It is immediate, your files are right there, and you are in the loop for every step.
Send unattended and scheduled jobs to the cloud
For long or recurring work, run it in the cloud so it does not depend on your laptop. A scheduled routine, an overnight enrichment across the whole list, these belong where they can run without you present.
TipThe test is simple: if the job should finish whether or not your laptop is awake, run it in the cloud.
Guard anything that runs without you
Before an unattended run, set the fence: permissions that keep it from destructive or send actions, validation on the files it produces, and an alert when it finishes. An unattended job needs guardrails a supervised one does not.
Start and approve from your phone
Use the mobile app to kick off or check on a run from anywhere, so a cloud job is one you approve from the elevator and a desk job is one you nudge from the hallway:
"How far is the overnight enrichment, and is anything waiting on my approval?"
Frequently asked questions
Is cloud Claude Code different from local?
No, it is the same agent with the same capabilities and your same setup. What differs is where the work runs, your machine or managed infrastructure, which maps to whether you are hands-on or the job is unattended.
When should I run locally?
For hands-on, iterative work where you are steering each step: tuning a prompt, working a messy file, building a skill. Local is immediate and keeps data on your machine.
When should I use the cloud?
For unattended and scheduled runs, a long enrichment, a Monday pipeline pass, so the job does not depend on your laptop staying awake. Anything that runs on a cadence belongs there.
Can I start a job and check it from my phone?
Yes. Remote control lets you kick off or check on a run from the mobile app, so you can approve a cloud job or nudge a desk job between meetings, without being back at your machine.
What about sensitive customer data in the cloud?
Be deliberate. Cloud runs happen on managed infrastructure, so match the data's sensitivity to where the job runs and check your organization's policy before pushing sensitive records through a cloud run.
What guardrails does an unattended run need?
Permissions that block destructive or send actions, validation on the files it produces, and an alert when it finishes. A job running with no one watching needs a fence a supervised one does not.
Is cloud more expensive than local?
Cloud runs on managed infrastructure you are billed for, while local uses your own machine. Most operators blend the two: local for hands-on work, cloud for long or recurring jobs, choosing per task.
How does this relate to Routines?
Routines are scheduled runs, which live in the cloud by nature, because the point is that they happen without you initiating them. See the Routines guide for setting those up.
Sources & further reading
Claude ships fast. This page was last reviewed Aug 22, 2026; verify time-sensitive details against the official docs above before relying on them.