Claude Routines are saved Claude Code configurations - a prompt, one or more repositories, an environment, and a set of connectors - that run automatically on Anthropic-managed cloud infrastructure on a schedule, an API call, or a GitHub event.
Anthropic shipped Routines in research preview in April 2026 as part of the redesigned Claude Code product family. They are not a separate product - they live inside Claude Code on the web alongside the interactive cloud sessions you already use. The pitch in the docs is direct: "Put Claude Code on autopilot." Source.
For marketers, that means a class of weekly research, audit, and roll-up tasks that used to require a dedicated automation tool can now sit inside the same Claude account that already drafts, edits, and audits your content - if you have the work in a GitHub repo Claude can clone.
What Claude Routines Are NOT
Anthropic ships several distinct products under the Claude name. Routines is a primitive inside Claude Code, not one of them. Get this wrong and the rest of your stack decisions go sideways.
Routines is not Claude Cowork. Cowork is the consumer Desktop autonomous-work surface for non-coding tasks (documents, spreadsheets, scheduled checks, file moves). Cowork is its own product. Routines lives inside Claude Code.
Routines is not Claude Managed Agents. Managed Agents is the developer cloud-hosted agents API on the Claude Developer Platform, built for embedding into your own product. Routines is the in-Claude-Code scheduling primitive. Different audience, different surface.
Routines is not Claude Design. Claude Design is the conversational design product for one-pagers, decks, and landing-page concepts. Different launch, different surface, no overlap with Routines.
Routines is not n8n or Make.com. Those are dedicated automation platforms with hundreds of integrations and a visual workflow canvas. Routines is narrower: scheduled execution of a Claude Code session against your repo. The honest comparison is later in this guide.
Where Claude Tasks Run: Three Scheduling Tiers
Claude Code offers three ways to schedule recurring or one-off work. Knowing which one to reach for is half the battle. Routines is the cloud tier - the rest of this page assumes Routines unless explicitly noted. Source.
Cloud Routines
Runs on: Anthropic cloud
Run on Anthropic-managed infrastructure. Work when your laptop is closed. Minimum interval one hour. Available on Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled. This page is about these.
Desktop Scheduled Tasks
Runs on: Your machine
Run on your machine. Persist across restarts but require the machine to be on. Have access to local files. Created in the Claude desktop app under Routines > New routine > Local. Minimum interval one minute.
/loop in CLI
Runs on: Your machine, current session
Session-scoped. Lives in the current Claude Code conversation and stops when you start a new one. Use for quick polling, watching a build, or babysitting a PR while the session stays open. Minimum interval one minute.
How Routines Actually Work
A Routine is created from one of three places: the web at claude.ai/code/routines, the desktop app under Routines > New routine > Remote, or the CLI with the /schedule command. All three write to the same cloud account, so a Routine created in any surface shows up in the others immediately.
The form sets up six things: the Routine name, the prompt that runs each time, one or more GitHub repositories Claude clones at the start of every run, the cloud environment (network access, env vars, setup script), the connectors Claude can use (MCP servers like Slack, Linear, Google Drive), and one or more triggers. A single Routine can have any combination of triggers.
Quick CLI examples
/schedule daily PR review at 9am /schedule tomorrow at 9am, summarize yesterday's merged PRs /schedule in 2 weeks, open a cleanup PR that removes the feature flag /schedule list # see all routines /schedule update # change one /schedule run # trigger immediately
Three trigger types are available, and you can attach any combination to the same Routine:
Recurring cadence (hourly, daily, weekdays, weekly) or a one-off run at a specific timestamp. Times are entered in your local zone. Minimum interval is one hour. Custom cron expressions are supported via /schedule update in the CLI.
A dedicated HTTP endpoint per Routine. POST with a bearer token to the /fire URL and a new session starts. The optional text field in the request body is passed to the Routine alongside the saved prompt as freeform context (an alert body, a campaign ID, a failing log line). The endpoint ships under the experimental experimental-cc-routine-2026-04-01 beta header and the docs warn shapes may change.
React to repository events. Supported categories during research preview: pull request and release, with action-level filters such aspull_request.opened or pull_request.closed. Field filters cover author, title, body, base/head branch, labels, draft state, and merged state. Each matching event starts its own session - no session reuse across events. Per-Routine and per-account hourly caps apply.
Each run is a full Claude Code cloud session: shell access, Skills committed to the cloned repo, every connector you included. By default Claude can only push to branches prefixed with claude/. Loosen this per-repository with the Allow unrestricted branch pushes permission if your prompt needs to update a real branch directly.
Are Routines Free? Plan Availability and Limits
No. Routines are not part of the Claude free tier. They are available on Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled. Source: Anthropic Routines docs.
Three things to know before you turn a daily report into a Routine:
- Each run draws down your subscription usage exactly like an interactive Claude Code session would. A Routine that runs every weekday is roughly five sessions a week against your plan.
- Each account has a daily cap on how many Routine runs can start. The exact number is shown at claude.ai/code/routines and claude.ai/settings/usage. One-off runs are exempt from this cap (they still consume regular usage).
- Organizations with extra usage enabled can keep running Routines on metered overage when the daily cap or subscription usage limit is hit. Without it, additional runs are rejected until the window resets.
6 Marketing Automation Use Cases for Claude Routines
The Anthropic docs lead with engineering examples (PR reviews, deploy verification, library ports). The same trigger primitives work for marketing operations as long as the work touches a repo Claude can clone. The recipes below are six that map cleanly onto the docs and onto the work most marketing teams already do badly.
Daily Content Brainstorm at 9am
A weekday-morning Routine reads your topic backlog plus yesterday's top-performing posts, returns three on-brand ideas with hooks and target platforms, and posts them to Slack before standup.
When to use
You burn the first 30 minutes of every morning trying to remember what to post about. Make Claude do that part before you open your laptop.
Trigger
Schedule trigger - weekdays at 9am local. (Pick a minute that is not :00 or :30 to avoid the one-shot jitter range described in the Anthropic docs.)
Routine prompt
Read references/topic-backlog.md and references/last-week-top-posts.md. Return three on-brand content ideas for today, each with: a 1-line hook, target platform, target persona, and the angle from the backlog it builds on. Post the result to the #content-ideas Slack channel.
Weekly Competitor Monitoring
Every Monday morning, the Routine clones your competitor-tracking repo, fetches each competitor's latest posts and pricing changes via the connectors, diffs against the prior week, and opens a PR with a markdown digest.
When to use
Your team is supposed to "watch the market" but nobody actually does it. Move it to a Routine that ships a digest with your name on the commit.
Trigger
Schedule trigger - weekly, Monday at 8am local.
Routine prompt
For each competitor in references/competitors.md, fetch their last seven days of LinkedIn and X posts plus their current pricing page. Diff against last week's snapshot in data/last-week.md. Return a markdown digest grouped by competitor with: new posts (link + 1-line summary), pricing changes, and three things we should respond to this week. Save the digest to data/digest-YYYY-MM-DD.md and open a PR.
Monthly Content Audit + Report
A monthly Routine pulls every published post from the last 30 days, scores each against your brand voice rules, ranks by engagement, flags content that broke voice, and ships a one-page report.
When to use
You know quality drifts when nobody is checking, but reading 200 posts a month is unrealistic. The Routine does the reading.
Trigger
Schedule trigger - first of every month at 7am local. Use a custom cron expression via /schedule update for monthly cadence.
Routine prompt
Read every post in posts/published/*.md modified in the last 30 days. For each post: score against the rules in references/brand-voice.md (1-10), record engagement from the analytics export at data/engagement-export.csv, and flag any post scoring below 6. Return a one-page report with: top 5 posts by engagement, bottom 3 voice scores with quote-and-fix, and three patterns from this month worth keeping or killing. Save to reports/audit-YYYY-MM.md and open a PR.
Brand Voice Drift Detection
A Routine runs on every PR opened against your content repo. It compares the new post to your brand voice rules and the last 20 published posts, then leaves inline review comments on lines that broke voice.
When to use
You hired a writer or another AI tool is drafting copy. You want a sanity pass before anything ships, without you reading every draft yourself.
Trigger
GitHub trigger - pull_request.opened on the content repo. Filter to PRs that touch posts/**.md.
Routine prompt
Read every changed .md file in this PR. For each file, compare against references/brand-voice.md and the last 20 posts in posts/published/. Leave inline GitHub comments on lines that break a rule, quoting the rule and suggesting a rewrite. Add a top-level comment with the overall verdict: ship as-is, ship with edits, or send back for rewrite.
Campaign Performance Summary
Your CD pipeline or analytics tool POSTs to a Routine's API endpoint at the end of every campaign. The Routine pulls the campaign data, calculates ROI against your benchmarks, and posts a one-screen summary to Slack.
When to use
You want a definitive "did this work?" answer the moment a campaign ends, not three days later when somebody finally pulls the report.
Trigger
API trigger - HTTP POST to the Routine's /fire endpoint with the campaign ID in the request body. (Authenticated with a per-Routine bearer token; the docs warn the experimental beta header may change.)
Routine prompt
Read the campaign ID from the request body. Pull the campaign's impressions, clicks, conversions, spend, and revenue from the analytics export. Compare against the benchmarks in references/campaign-targets.md (CPA, ROAS, CTR). Return a Slack message with: campaign name, headline metric (ROAS), three wins, three misses, and a ship-it-again / iterate / kill verdict. Post to the #marketing-results channel.
Weekly Newsletter Roll-Up
Every Friday, a Routine reads your week's published posts, top engagement from analytics, and three saved articles from your reading list, then drafts a newsletter in your voice and opens a PR with the draft.
When to use
You ship a newsletter every week and the prep keeps slipping to Friday afternoon. Move the drafting to a Routine and edit instead of starting from a blank page.
Trigger
Schedule trigger - Friday at 9am local.
Routine prompt
Read every post published this week in posts/published/. Pull top three by engagement from data/engagement-export.csv. Read three latest items in reading-list.md. Draft this week's newsletter in the voice defined in references/brand-voice.md, with: a 1-line opener, the three featured posts (1 paragraph each, link), one external recommendation, and a clear CTA. Save to drafts/newsletter-YYYY-MM-DD.md and open a PR.
Claude Routines vs n8n vs Make.com
The honest comparison. Routines is younger, narrower, and Claude-native. n8n and Make.com are mature general-purpose automation platforms with broad integration catalogs. Pick on the shape of the work, not the brand.
| Dimension | Claude Routines | n8n | Make.com |
|---|---|---|---|
| Where it runs | Anthropic cloud | Your server, n8n cloud, or desktop | Make.com cloud |
| Workflow style | One Claude Code session per run, full reasoning | Visual node graph; AI nodes optional | Visual scenario graph; AI modules optional |
| Integrations | MCP connectors + whatever your environment can call | Hundreds of native nodes; broad self-host ecosystem | Hundreds of native modules |
| Triggers | Schedule, HTTP API, GitHub events | Schedule, webhook, polling, app-specific | Schedule, webhook, polling, app-specific |
| Minimum interval | 1 hour | Sub-minute (your infra) | 1 minute (paid plans) |
| Pricing model | Bundled in Claude Pro/Max/Team/Enterprise | Free self-host; cloud plans on n8n.io/pricing | Free tier + paid plans on make.com/pricing |
| Best for | Recurring tasks where a Claude session is the work | Multi-tool data plumbing with self-host control | Multi-tool data plumbing without self-host |
| Worst for | Workflows that are mostly tool-to-tool data shuffling | Tasks where reasoning is the whole point | Tasks where reasoning is the whole point |
Reach for n8n or Make.com when the value lives in stitching tools together: pull from Stripe, format in JSON, write to Airtable, send to Slack, log to a dashboard. The visual canvas pays for itself the moment the workflow grows past three steps and three apps.
Reach for Claude Routines when the work itself is Claude doing the thinking on a recurring cadence: read a draft and audit voice, diff this week's competitor posts and write a digest, score yesterday's campaign and flag misses. n8n can call Claude as a node, but you spend your week stitching graphs around what is essentially a Claude prompt that wants to be the whole workflow.
Many marketing teams will use both. n8n or Make.com for cross-tool data movement. Routines for the Claude-shaped slice. The two layers do not compete - they cover different parts of the stack.
How Brand Brain Uses Scheduling (and Where Routines Fit)
Brand Brain is a content operations platform with scheduled posting baked in. You approve a draft, choose the platform mix, set a publish window, and Brand Brain pushes the post to LinkedIn, X, Instagram, and Threads at the scheduled time. TikTok and YouTube support are on the roadmap (coming soon). The scheduling job lives inside the Brand Brain workflow, not outside it.
Honest positioning
Routines and Brand Brain scheduling solve different problems. Routines is for ad-hoc Claude Code tasks running on a cadence (audits, diffs, roll-ups, drafts against a repo). Brand Brain scheduling is for the always-on social publishing pipeline: approved content going out at the right time on the right platforms with the right voice.
Routines vs Brand Brain Always-On Agents
Brand Brain ships its own scheduled-agent layer called Always-On Agents — proactive AI agents that run on a cadence you set without you opening the app. The dashboard already includes Trend Detection, Auto Content Reviewer, and LinkedIn Product Promotion agents, plus custom schedules with ON/OFF toggles. They scan trends, draft content, and queue results for your review while you sleep. This is the closest like-for-like comparison to Routines on the BrandBrain side.
| Dimension | Claude Routines | BrandBrain Always-On Agents |
|---|---|---|
| Shape of work | General Claude Code session against your repo on a schedule | Marketing-shaped agents pre-wired for content ops |
| Setup | Write a prompt, point at a repo, attach connectors and triggers | Toggle pre-built agents ON or define a custom one in the dashboard |
| Pre-built tasks | None — every Routine starts blank | Trend Detection, Auto Content Reviewer, LinkedIn Product Promotion, custom schedules |
| Repo requirement | Yes — each run clones a GitHub repo | No — runs against your BrandBrain workspace and brand voice |
| Output destination | Markdown file, PR, Slack message, or whatever your prompt writes | Review queue in the dashboard or Telegram approval |
| Best for | Audits, diffs, roll-ups against a repo (engineering-shaped work) | Daily content drafts, trend scans, voice-checked queue items |
Pick Routines when the work is Claude reasoning against a repository (content audit, brand-voice diff on a PR, weekly competitor digest committed to a tracking repo). Pick BrandBrain Always-On Agents when the work is a recurring marketing job that needs to land in a review queue and ship to social — trend scanning into your draft queue, auto-reviewing inbound content, drafting LinkedIn product promotions on a cadence. The two layers cover different surfaces of the same problem and most teams using both get the cleanest split.
A realistic stack for a marketing team using both:
- Claude Routines for the weekly competitor digest, monthly content audit, brand voice drift detection on every PR, and campaign performance roll-ups (the recipes earlier on this page).
- Brand Brain for the daily content engine: voice training, draft generation, review queue, multi-platform publishing, and the schedule itself.
- n8n or Make.com for the cross-tool plumbing if your stack has enough non-Claude, non-publishing steps to need a graph.
The trap is using a Routine for what should be a Brand Brain post and vice versa. Routines are great at producing markdown, opening PRs, and posting to a Slack channel. They do not publish to LinkedIn, generate platform-native carousels, manage a content review queue, or train a brand voice across thousands of past posts. That is the platform layer.
Why This Matters Right Now
Routines launched in research preview in April 2026. Practitioners on YouTube are already running the "here is how I scheduled Claude Code to do X overnight" playbook. Nick Saraev demonstrated marketing automation use cases for Routines in his April 2026 walkthrough "Claude Routines Just Dropped".
If you already run Claude Code for content audits or research and you have a repo Claude can clone, the cost of moving one weekly task to a Routine is roughly the 15 minutes it takes to write the prompt. The cost of leaving the same task as a manual habit is paid every week, forever.