How to Run an AI Agent on a Schedule With Your Laptop Closed

Construct, an AI agent platform, published a guide showing how to run scheduled AI agent tasks in its cloud workspace so jobs fire with the laptop closed, with plans priced at $9 a month for 3 scheduled tasks, $59 for 10, and $299 for 50. The guide contrasts this with self-hosted agents such as OpenClaw, whose cron jobs fire only while the Gateway process runs on a machine that stays on, and with Claude Cowork scheduled tasks, which run remotely unless a task needs local files or apps, per the Claude Help Center. Construct's comparison table, checked October 11, 2026, lists a 15-minute shortest repeat for agent jobs and 1 minute for notifications and app tools, and the guide walks through building one recurring weekday 8am client follow-up job.

How to Run an AI Agent on a Schedule With Your Laptop Closed Nischal https://construct.computer/authors/nischal/ · Author Writes about Construct, AI agents, and building useful software. Published · @naik nischal https://x.com/naik nischal Construct runs AI agent scheduled tasks in the cloud, so they fire with your laptop closed: Lite at $9 a month includes 3 scheduled tasks, Starter at $59 includes 10, and Pro at $299 includes 50. Self-hosted agents such as OpenClaw run automations "inside the Gateway process," and "the Gateway must be running for schedules to fire" OpenClaw docs https://docs.openclaw.ai/automation/cron-jobs/how-it-works , which is why people leave a Mac mini switched on in a cupboard. For founders and small teams, Construct is the simpler way to put an agent on a schedule. The agent's files, connected apps, browser, and terminal all live in its cloud workspace, every run keeps a history you can check from your phone, and finished work is saved as it is produced. This guide sets up one real recurring job, a weekday 8am client follow-up list, from the first on-demand run to a schedule you never have to babysit. Why do this in Construct? - Scheduled tasks on every plan, from $9. Lite includes 3, Starter 10, and Pro 50. Only active schedules count, so a paused job frees its slot. - Nothing on your desk has to stay awake. Calendar schedules run server-side, and the files, apps, browser, and terminal a job needs are part of the agent's cloud workspace, so a run never reaches back to your machine. - You decide what happens when a run is late. Delivery behavior lets you skip or queue an overlapping run, and run once, skip, or catch up safely on a late one. - A history for every run. Each Calendar event lists its recent runs with the outcome and a link to the chat, readable from the web desktop on your phone. - Finished work survives a stopped run. Files are saved as the agent produces them, so a run that stops at step 30 keeps steps 1 to 29. - It grows into a workflow. Once a job settles, save it as a workflow with app steps and notifications, and schedule that instead. Why do scheduled AI agent jobs stop when the laptop closes? They stop because the scheduler, or something the job needs, lives on your machine. A job runs with the lid shut only if the scheduler, the files, the logins, and the connected apps all live in the cloud too. Even cloud schedulers carry that catch. Claude's Cowork scheduled tasks "run remotely, so they run on their cadence even when your computer is asleep," but "if a scheduled task requires local files or apps, it will only run locally" Claude Help Center https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork . Here is how the common options compare, checked October 11, 2026. | Where AI agent scheduled tasks run, checked October 11, 2026 | | | | |---|---|---|---| | Option | Runs with your laptop closed? | Shortest repeat | Price to start | |---|---|---|---| | Construct Calendar | Yes; the schedule, files, apps, browser, and terminal are all hosted | 15 minutes for agent jobs; 1 minute for notifications and app tools | $9 a month, 3 scheduled tasks | | OpenClaw cron jobs | Only if the gateway runs on a machine that stays on, such as a VPS | Set by your cron expression | $0 software, plus machine and tokens | | Claude Cowork scheduled tasks | Yes, unless the task needs local files or apps | Hourly | Any paid Claude plan | | Claude Code routines | Yes, on Anthropic's cloud; research preview, built around code repositories | One hour | Pro, Max, Team, or Enterprise | Sources: Construct pricing https://construct.computer/pricing/ , OpenClaw cron docs https://docs.openclaw.ai/automation/cron-jobs/how-it-works , OpenClaw hosting FAQ https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting , Claude Help Center https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork , and Claude Code routines https://code.claude.com/docs/en/routines . OpenClaw's own FAQ says to "expect pauses when the machine sleeps or updates" and that 24/7 reliability means a VPS OpenClaw FAQ https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting . In Construct, a hosted workspace is the default; our explainer on AI agents with their own computer https://construct.computer/blog/ai-agent-with-its-own-computer/ covers why it matters. Is an AI agent schedule the same as a cron job? Not quite. A cron job runs the same fixed command every time; a scheduled agent job runs a prompt, and the agent decides the steps on each run. That flexibility is the point, because the agent can cope with a changed page or an extra email, but it means the prompt has to carry the rules a cron script would hard-code: what to read, where to save, and what never to do. Construct takes the timing side off your hands. You pick presets such as Every weekday instead of writing cron syntax, the event keeps its own time zone, and missed or overlapping runs follow the delivery rules you set in step 3. What do you need? - A Construct plan with a free scheduled task. Lite includes 3, Starter 10, Pro 50. Only active schedules count; a paused one does not. There is also a 7-day Pro trial. See pricing https://construct.computer/pricing/ . - One job that has already worked on demand. Schedule a job you have run and checked at least once, not one you are describing for the first time. - Any apps the job touches, connected first. For this example, Gmail and Google Calendar, connected from Construct's integration catalog. A schedule rechecks the connection every time it fires, so connect before you schedule. - Your time zone set in your profile. New schedules use it by default. Before you start - Agent jobs repeat at most every 15 minutes. Schedules that only send a notification or run an app tool can repeat every minute. For anything that must react the moment it happens, forward it to the agent's own email address https://construct.computer/blog/ai-agent-email-address/ : mail from your own address starts a run without waiting for a schedule. - Write jobs that check before they act. Construct delivers a scheduled run at least once, not exactly once, so the prompt in step 2 stops if today's file already exists. - Keep sends out of the job until it has proved itself. Scheduled runs do not pause for approval, so ask for drafts or files and send anything external yourself. - Size the job to your plan. Each run is limited by steps per task: 50 on Lite, 150 on Starter, and up to 1,000 on Pro. Split a long job into two scheduled tasks, or move up a plan. - Construct's Calendar sets the schedule. Your connected Google Calendar is something the agent can read and update, not where schedules live. How do you set up AI agent scheduled tasks in Construct? You run the job once while you watch, ask the agent to put it on the Calendar, set what happens when a run is late, and test it with Run now . The whole setup takes one conversation and one look at the event editor. 1. Run the job once while you watch Ask for the job in chat exactly as you would want it done at 8am: Open the file and check it. If the list missed something, say what, and run it again. Fix the prompt now, while you are watching, not after it has run unattended for a week. 2. Ask the agent to put it on the Calendar Once the output is right, schedule it from the same chat: The agent creates a Calendar item it owns, with the prompt as the action. You can also build it by hand: open Calendar, press New event , give it a title, set Repeats to Every weekday , choose the Owning agent , and pick Ask agent under "What should happen?" with your prompt as the input. The last sentence of that prompt is not decoration. Construct delivers a scheduled run at least once, not exactly once, so a job that checks before it writes is safe if it ever fires twice. 3. Set what happens when a run is late or still going In the event editor, open Delivery behavior . Two choices matter: - If still running: Skip overlap drops the next run if the last one has not finished; Queue next waits and runs it after. For a daily list, skip. - If delivery is late: Run once runs a late occurrence once, Skip drops it, and Catch up safely replays missed runs oldest first, in small batches. For a morning list, run once: you want today's list, not three stale ones. 4. Test it with Run now Open the event and press Run now . This fires one standalone run without moving the schedule. Check the file it wrote, then close the laptop and wait for 8am. 5. Check the history from anywhere Open the event and look at Recent fires . Each entry shows when the run was due, its outcome, the number of attempts when there was more than one, and, for agent prompts, a link to the chat session the run used. You can read it from the web desktop on your phone; nothing about checking a run needs the laptop. 6. Promote it to a workflow when it settles After a couple of weeks of good runs, save the job as a workflow and schedule the workflow instead, using Run workflow as the Calendar action. A workflow can add a connected-app step after the agent step, such as saving the list as a Notion page, plus an in-app notification, and published versions stay fixed so a later edit does not change a run already in progress. Our workflow guide https://construct.computer/blog/ai-workflow-automation/ and the workflows overview https://construct.computer/use-cases/workflows/ cover that step. What result should you get? After the first scheduled morning you should have four things, none of them on your laptop: 1. A dated file in Files/follow-ups/ with the client list and today's meetings. 2. A Calendar event set to Every weekday , with its next run shown. 3. An entry under Recent fires , marked as succeeded or failed. 4. The chat session the run used, so you can read what the agent did and why. If a run dies partway, finished work is still there. A run that stops at step 30 keeps steps 1 to 29, and tomorrow's run starts from what exists. After the first week, check four things yourself: that Recent fires shows one entry per weekday, that Files/follow-ups/ has one file per day and no duplicates, that each run's chat shows no send or reply, and that the list still matches what you would have written by hand. If all four hold, leave it alone. Illustrative example. A solo bookkeeper schedules three jobs on Lite: the weekday follow-up list above, a Monday 7am summary of last week's invoices from Files, and a monthly reminder to chase late payers. That uses all three of Lite's scheduled tasks. A fourth job means pausing one or moving to Starter. Troubleshooting "Scheduled task limit reached." You have used your plan's scheduled tasks. Pause or cancel one, or upgrade. Paused events do not count. "Won't fire until scheduling is fixed." The event saved but did not arm. Press Retry on the warning, or wait: Construct heals unarmed schedules on its own within a few minutes. The schedule was rejected as too frequent. It runs an agent more often than every 15 minutes. Use 15 minutes or longer. It ran at the wrong hour. The event uses the time zone it was created with. Changing your profile time zone later does not rewrite existing events, so edit the event's time zone directly. A run stopped and asked you to connect an app. The app was disconnected or its sign-in expired. Disconnecting an app does not pause the schedules that use it; each run checks the connection when it fires. Use the connect card in the run's chat, or reconnect from Settings, Connections, then press Run now to confirm. Two files appeared for one day. A run was delivered twice. Add the "check whether today's file already exists" line from step 2. A resumed job did not catch up on the paused days. That is by design: a resumed recurring job runs at the next slot after now. Ask the agent to run the missed days by hand if you need them. You need to stop a run in progress. Use Stop all in the menu bar's live activity panel. It stops agent turns and workflow runs already in progress; pause the event to stop future runs. Already running scheduled jobs on your own machine? Move one in 15 minutes Start with the job that breaks most often when your laptop sleeps, such as an OpenClaw cron job or a local desktop task. 1. Start on Lite for $9, or take the 7-day Pro trial if the job needs more than 50 steps. 2. Paste the job's instructions into a Construct chat, and move the files it reads into the workspace. 3. Connect the apps it touches from the integration catalog. 4. Run it once and check the file it saves, as in step 1 above. 5. Ask the agent to schedule it , then press Run now and close the laptop. There is no import, so move one job at a time. A job that reacted to incoming mail becomes a scheduled inbox check, or an email you forward to the agent's own address. If you are weighing a managed OpenClaw host instead, our guide to hosted OpenClaw alternatives https://construct.computer/blog/hosted-openclaw-alternatives/ compares them. Give Construct its first scheduled job Choose Construct for scheduled agent work. It includes scheduled tasks from $9, runs them in the cloud with nothing left switched on at home, keeps every finished file and run history where your team can see it, and lets you decide what happens when a run is late. The scheduling page https://construct.computer/use-cases/scheduling/ shows how it fits the rest of your week. Your first job is the one above: run the follow-up list prompt from step 1 today, fix it while you watch, then schedule it for 8am every weekday and close the laptop. Tomorrow morning, open Files/follow-ups/ on your phone. Start on Lite at $9 https://construct.computer/pricing/ , or try Pro free for seven days. Give Construct its first scheduled job https://app.construct.computer Sources Pages checked on October 11, 2026: Related resources Frequently asked questions - Can an AI agent run scheduled tasks with my laptop closed? - Yes, if the scheduler and everything the job needs live in the cloud. Construct's Calendar runs schedules on its servers, and the agent's files, connected apps, browser, and terminal are part of its cloud workspace, so a run never reaches back to your machine. Lite at $9 a month includes 3 scheduled tasks, Starter at $59 includes 10, and Pro at $299 includes 50. - How often can a scheduled AI agent job run in Construct? - A schedule that runs an agent, or a workflow with an agent step, can repeat at most every 15 minutes. Schedules that only send a notification or run an app tool can repeat every minute. For something that must react the moment it happens, forward it to the agent's own email address, since mail from your own account address starts a run without waiting for a schedule. - Do Claude's scheduled tasks run when my computer is off? - Claude's Cowork scheduled tasks run remotely, so they keep their cadence while your computer is asleep, but a task that needs local files or apps only runs locally. Claude Code routines run on Anthropic's cloud with a one-hour minimum interval and are in research preview. In Construct, the files and apps a job needs already live in the hosted workspace, so nothing depends on your laptop. - Why do OpenClaw cron jobs stop running? - OpenClaw runs automations inside its gateway process, and its docs say the gateway must be running for schedules to fire. If the gateway is on a laptop, expect pauses when the machine sleeps or updates; OpenClaw's own FAQ points to a VPS for 24/7 reliability. Construct runs schedules on its own servers, so there is no machine of yours to keep awake. - Is an AI agent schedule the same as a cron job? - Not quite. A cron job runs the same fixed command every time, while a scheduled agent job runs a prompt and the agent decides the steps on each run. That lets it cope with a changed page or an extra email, but the prompt has to carry the rules a script would hard-code: what to read, where to save, and what never to do. - What happens if a scheduled run is late or the last one is still running? - In the event's Delivery behavior settings you choose Skip overlap or Queue next for a run still in progress, and Run once, Skip, or Catch up safely for a late one. Catch up safely replays missed runs oldest first, in small batches. For a daily list, skipping overlaps and running a late occurrence once is a sensible default. - Can a scheduled agent run fire twice? - Occasionally. Construct delivers scheduled runs at least once, not exactly once, so write jobs that check before they act. For a daily report, add a line telling the agent to stop if today's file already exists. That one sentence makes a duplicate delivery harmless. Keep reading AI Workflow AutomationCreate reusable linear AI workflows, run them on demand, or schedule recurring work across files, live browser runs, native email, and connected business apps. Agent reliability calculator: how many steps will your agent finish?AI agent reliability calculator: a 95% reliable agent finishes a 48-step job in one run 8.5% of the time. See retries, half-life, and what checkpoints fix. Your agent has a half-lifeWhy AI agents keep failing on long multi-step jobs: a 95% reliable agent finishes 48 steps 8.5% of the time. The fix is a resumable run, not a better model. Grok Bot Alternative: Construct vs Grok BotCompare Construct and Grok Bot on monthly pricing, workflows, integrations, memory, and model choice for small teams.