Cowork Scheduled Tasks: The Workflow Layer Most Plugins Aren't Using Yet
Cowork scheduled tasks turn Claude into a recurring background coworker. Three concrete profession-specific workflows you can ship this weekend.
A Claude that runs without you in the room is a different product than a Claude you chat with.
That sounds obvious. It isn't. Almost every "AI workflow" article shipped in the last 18 months treats Claude as something you talk to: open a tab, paste a prompt, copy the output. Cowork's scheduled tasks invert that. The agent runs on a cron. It reads files you've left in a folder. It produces a deliverable that's waiting when you sit down. The model is the same; the surface is not, and the surface is what makes a workflow.
This piece is for people building real profession-specific Claude setups — agents, plugins, vault buyers, power users. Three workflows worth setting up this weekend, plus what changed in April 2026 that makes templated scheduled tasks a viable product category.
What scheduled tasks actually are
Scheduled tasks are a Cowork feature that runs a Claude agent on a recurring or one-off schedule. They have three properties that make them load-bearing for profession-specific work:
- Cron-style recurrence. "Every Monday at 7am Pacific." "Last business day of every month at 6pm." "Every weekday at 4:55pm." The agent runs without a human prompting it.
- Persistent agent thread. Each scheduled task lives inside a thread you can open from mobile, desktop, or wherever Cowork is — so the run history is reviewable, the agent's reasoning is auditable, and you can fork to a manual conversation when something goes off-script.
- Customize menu integration. Scheduled tasks can pull from the same skills, plugins, and connectors you'd use in a manual session. So a task can call your Pharmacist plugin's
/draft-prior-authslash command, read from a Google Drive connector, and post to Slack — without bespoke glue code.
The new ingredient that makes 2026 different is the 1M context window going GA on March 13 for Sonnet 4.6 and Opus 4.6 at standard pricing — including 600 images or PDF pages per request, up from 100. A scheduled task can now chew through a full week of clinical PDFs, transaction exports, listing photos, or session-note JSONs in a single run. That changes which workflows are realistic to template.
Three concrete workflows
Workflow 1 — Real estate agent: Monday 7am pipeline review
A solo agent with 18 active buyers and listings drops their weekly Trello-to-CSV export into a Google Drive folder every Sunday night. They want to walk into Monday morning with one document that says: where the deals are at risk, which buyer agreements expire this week, and what to send the top three contacts before the 9am calls.
The scheduled task config:
Schedule: 0 7 * * 1 (Mondays, 7am local)
Plugin context: real-estate@alexclowe/awesome-claude-cowork-plugins
Connectors: Google Drive (read-only on /weekly-pipeline)
Skills: pipeline-review, monday-outreach-draft
Output: Markdown to /weekly-summary, push notification on critical riskWhat it does:
- Reads the latest
pipeline-YYYY-MM-DD.csvin the folder. - Joins buyer agreement dates from the last project state file.
- Flags deals that haven't progressed in 14+ days, expiring agreements, and price reductions on competing comps.
- Drafts three outreach messages — fair-housing-compliant, voice-matched to the project's saved tone — for the highest-priority buyers.
- Writes a summary file the agent reviews with their morning coffee. Pushes a notification only if a deal is flagged "critical."
This is the same kind of workflow that ships in The Real Estate Agent Claude Vault — a packaged Claude Project setup with the project file, the prompts, and the fair-housing guardrails that make a recurring task safe to leave running.
Workflow 2 — Accountant: Last business day of month, 6pm month-end close digest
A bookkeeping firm with 22 monthly clients runs the same close ritual every month: pull each client's QuickBooks export, write a one-page close summary, flag variances over 10% versus the trailing three-month average, queue the client emails for first-of-month delivery.
The scheduled task config:
Schedule: 0 18 L-1W * * (last business day, 6pm)
Plugin context: bookkeeper@alexclowe/awesome-claude-cowork-plugins
Connectors: Google Drive or Dropbox (read on /clients/<slug>/exports)
Skills: month-end-close-summary, variance-flag, client-email-draft
Output: One PDF per client to /close-packets/<month>/What it does:
- Iterates through the configured client list.
- For each client, reads the most recent QuickBooks export plus the last three months' close packets.
- Produces a one-page summary: P&L vs. budget, top three variances, cash-position note, items requiring client confirmation.
- Drafts the first-of-month client email referencing the summary, in the firm's voice.
- Stops short of sending — humans review, click send.
The output is structured and reviewable, the variance threshold is configurable per client, and the agent thread is open if a partner wants to ask follow-ups before the client meeting. The Bookkeeper plugin is the natural starting point if you're building this from scratch.
Workflow 3 — Therapist: Friday 5pm caseload summary
A solo therapist with 28 active clients keeps structured session notes in a private file system the rest of the week. On Friday afternoon they want a one-pager per client: progress against last week's plan, risk flags worth a Sunday-night review, plan-of-care notes for the coming week, and a flagged list of insurance authorizations expiring in the next 14 days.
The scheduled task config:
Schedule: 0 17 * * 5 (Fridays, 5pm local)
Plugin context: therapist@alexclowe/awesome-claude-cowork-plugins
Skills: caseload-summary, risk-flag, auth-renewal-batch
Output: One markdown file per client to /caseload-summaries/<week>/What it does:
- Reads the week's session notes, indexed by client.
- For each client, summarizes progress, scores risk against a configurable rubric, drafts the next-week plan.
- Builds a single auth-renewal worklist for clients whose insurance authorization expires in the next 14 days.
- Pushes a notification only if a risk flag exceeds threshold — the rest is quiet output the therapist reads on Sunday before opening Monday's first session.
Risk-flag thresholds, the auth-renewal lookahead window, and the note style are all project-level settings, not buried in the prompt. The Therapist plugin ships with the slash commands and skills this kind of task hooks into.
What changed in April 2026
Three quiet shipping decisions over the last 60 days are why scheduled-task templates went from "interesting" to "viable product category":
- Mature plugin marketplaces with skills, agents, hooks, and MCP servers. The official Claude Code plugin marketplace and the third-party marketplaces around it now ship pre-configured plugin packages — including profession-specific ones — that scheduled tasks call as first-class capabilities. Discoverability and install friction stopped being the bottleneck.
- Push notifications via Remote Control. Per the April 2026 Claude Code release notes, version 2.1.110 added a push-notification tool that lets Claude send mobile pushes when Remote Control and the "push when Claude decides" config are enabled. A scheduled task that finds something worth a human's attention can ping the user's phone instead of writing into a folder no one will check. The same release also added
--resume/--continuefor unexpired scheduled tasks, so you can pick up a recurring run after restart. - 1M context at standard pricing. Six-hundred PDF pages or images per request, billed at the same per-token rate as a 9K-token request, removes the throttle that used to make end-of-week or end-of-month batch tasks awkward. You can drop a week of session notes, a month of QuickBooks exports, or a quarter of listing photos into one run without splitting.
The combination — mature plugin distribution, attention-routing via push, and a context window large enough to chew through a real workload — is what makes these templates ship-able instead of demo-able.
Why this is the structural argument
A general-purpose chat tool, no matter how good the base model, can't run on a cron, hold project-level state across documents, or call profession-specific helpers without a plugin layer. Scheduled tasks are the workflow surface that lives outside chat. That's the structural difference between an AI you talk to and an AI that does the work — a distinction that matters for clinicians thinking through whether a free chat tool covers their use case, and for therapists thinking through what a profession-specific documentation workflow actually looks like end-to-end.
If you're a builder, the question is which scheduled-task template makes your plugin twice as valuable. If you're a buyer, the question is which one of the three workflows above maps to the after-hours work you'd give back if a coworker would do it for you.
Either way: scheduled tasks are the layer to be paying attention to in 2026.
Get started
- Browse free profession plugins and pick the one closest to your work — every plugin includes the slash commands and skills referenced above.
- The Real Estate Agent Claude Vault ships the Monday-pipeline workflow above as a complete Claude Project setup with project file, 50 prompts, and fair-housing guardrails — $29 one-time.
- The AI Readiness Audit is a free 5-minute scorecard that identifies which scheduled-task templates would save you the most time given your current workflow.
For builders shipping their own Cowork plugins, the Cowork Plugin Builder Guide walks through the plugin manifest, the skills system, slash commands, and how to expose helpers that scheduled tasks can call cleanly. The first plugin that ships a polished scheduled-task template per profession is going to define the category.
Related Guides
The 2026 PT Documentation Crisis: Why CMS TEAM + ACCESS Made AI Workflow a Survival Skill
TEAM is mandatory since January 2026. ACCESS launches July 2026. What the new CMS rules mean for PT documentation — and the AI workflow scribes miss.
AI Agents vs Chatbots for Professionals: What's the Difference and When You Need Each
Chatbots are reactive. Agents are persistent and can take actions. Here's the honest breakdown of which one a professional actually needs in 2026.
Best AI Tools for Attorneys in 2026
A curated list of the best AI tools for attorneys in 2026 — demand letters, contract review, client memos, billing narratives, and document drafting.