Automate the boring parts

Automate what you have already done ten times

The rule that saves most people from wasting a month: never automate something you have not done by hand at least ten times. The first ten runs teach you where the job actually breaks, which fields matter, and whether it is worth doing at all. Automating on run one means building a machine to do a job you do not yet understand.

By this step you have been publishing for a while, so you know which parts are dull and identical each time. Those are the candidates.

The three tools

All three connect apps to each other with triggers ("when a new article is published") and actions ("post to X, add a row to a sheet, send an email"). They differ in price model, ceiling, and how much they let you see.

Zapier

Zapier has the most integrations and the gentlest learning curve. The free plan gives 100 tasks a month with two-step Zaps only; Professional starts at $29.99 a month ($19.99 on annual billing) for 750 tasks and adds multi-step workflows, filters and webhooks. A task is billed each time an action runs, so a five-step Zap that fires daily costs five tasks a day. It gets expensive fast and it is the easiest to start with. Check the current page.

Make

Make shows your workflow as a visual map of modules, which makes complex logic (branches, loops, error handling) easier to understand than a list of steps. It bills in credits, one per module run; Core is $16 a month for 10,000 credits ($12 on annual billing), Pro $28. The free plan is enough to learn on. Make is where most creators land: cheaper than Zapier at volume, more capable, and still no code.

n8n

n8n is the technical choice and the one with no ceiling. Its source is open to read and free to self-host on a small server or your own machine; the cloud version starts from around 20 to 24 euros a month for 2,500 executions. It bills per workflow run, not per step, so a long workflow costs the same as a short one, which changes the economics of anything with many steps. It has native nodes for calling AI models, which is why it has become the default for AI pipelines. If you were comfortable with the plain-HTML route in step three, you will be comfortable with n8n.

Decision: Zapier to learn, Make to run, n8n to own.

The four automations to build first

One: the content pipeline

Trigger: a new row in a sheet or a new file in a folder marked "ready". Actions: send the draft to an AI model with your editing instructions from step four; generate a featured image with Flux through an API; create a draft post in your site with the title, body, image and category; message you with a link to review.

What it does not do: publish. The last click stays with you, every time. The pipeline's job is to remove the hour of copying, resizing and pasting, not the ten minutes of judgement.

Two: social posting

Trigger: a post is published (an RSS feed is the universal trigger here and every site from step three has one). Actions: ask the AI for three post variants in your voice with the article's key point, save them to a queue, and post one per day for three days, each linking to the newsletter page. Add a filter so only posts in certain categories fire.

Set it up and then read what it posted for the first two weeks. The AI will drift toward hype within days if you do not correct the prompt.

Three: the newsletter digest

Trigger: a schedule, every Thursday morning. Actions: collect the week's published posts from the RSS feed, collect the links you saved during the week (a bookmarking tool or a simple sheet), ask the AI to draft the digest in your template, and create a draft in Beehiiv or your platform of choice. You open the draft, rewrite it in your voice, and send. Drafting time drops from an hour to fifteen minutes and the rhythm holds.

Four: RSS monitoring

Trigger: new items in a set of RSS feeds (competitors, vendor changelogs, the subreddits and forums from step one, Google Alerts on your topic). Actions: filter by keywords, ask the AI to score each item for relevance to your readers on a one-to-five scale with a one-line reason, and drop anything scoring four or above into a channel or a sheet you check daily. This is the automation that most directly feeds a news site like ours. It replaces an hour of tab-opening with a five-minute scan of a pre-sorted list.

Build them in that order. Each one is a day of work with the free tier, and the fourth one is the one you will thank yourself for.

What not to automate

• Replying to people. A reader who emails, comments or replies to the newsletter is the most valuable person you will meet that day. An automated reply tells them they are not.
• Publishing. Everything can be drafted by a machine; nothing should go live without a human clicking.
• Fact-checking. A model checking a model is two guesses. The check in step four is a human job.
• Anything with money: sending invoices, approving refunds, changing prices. The blast radius of a bug is too large.
• Your voice. Automations can draft in a style. They cannot decide what you think. If a workflow ever produces a finished opinion piece, delete the workflow.
• Outreach. Automated "I loved your article" emails are spam and everyone can tell.

Keeping it from breaking

Every automation fails eventually, usually because an app changed its API or a field came back empty. Three habits:

• Put an error route in every workflow that messages you with the step that failed and the input it received.
• Keep a one-line log per run in a sheet: time, trigger, result. When something goes quiet, you will know when it stopped.
• Review the credit or task usage monthly. Runaway loops are the most common surprise bill in all three tools.

Checklist

• Listed the tasks done by hand at least ten times and picked the dullest
• Chosen Zapier, Make or n8n with the learn-run-own rule and read current prices
• Built the content pipeline with a human publish step
• RSS monitoring live with AI scoring and a daily review list
• Error notifications and a run log added to every workflow